Model lifecycle and updates

Type

Time

Deployment scope

Model

Description

Reference-to-video

2026-07-01

Chinese mainland

wan2.7-r2v-2026-06-12

A snapshot version of the Wan 2.7 reference-to-video model. It supports subject referencing and voice customization, and can generate a scripted video directly from a single, multi-panel storyboard image. Wan2.7 - reference-to-video

Image generation

2026-06-25

Chinese mainland

qwen-image-2.0-pro-2026-06-22

The latest snapshot of the Qwen-Image-2.0 series, unifying image generation and editing. Compared with the April 22 snapshot, this model features enhanced text rendering with support for up to 1k token instructions, more refined photorealistic quality and scene detail, and stronger semantic adherence.

Text-to-video

2026-06-22

Chinese mainland

happyhorse-1.1-t2v

HappyHorse 1.1 text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-06-22

Chinese mainland

happyhorse-1.1-i2v

HappyHorse 1.1 image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-06-22

Chinese mainland

happyhorse-1.1-r2v

HappyHorse 1.1 reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Text extraction

2026-06-16

Chinese mainland

qwen3.5-ocr

Qwen text extraction model based on the Qwen3.5 architecture, delivering faster speed and better performance. Context length extended to 128K with multi-turn conversation support. Information extraction capability significantly improved, covering various domestic and international ID cards and certificates.Text extraction (Qwen-OCR)

Reasoning model

2026-06-15

Chinese mainland

kimi/kimi-k2.7-code

Kimi K2.7 Code model from Moonshot AI, a coding-focused agentic model optimized for long-horizon software engineering tasks. Thinking-only model. Excels at multi-file refactoring, feature implementation, and long-session debugging workflows. Kimi-Moonshot AI

Reasoning model

2026-06-10

Chinese mainland

qwen3.7-max-2026-06-08

The Max model, the largest and most capable in the Qwen3.7 series, has added visual-modal understanding compared to the May 20 snapshot, enabling it to perceive real-world scenes and supporting multimodal interactive hybrid agent capabilities.

Reference-to-video

2026-06-08

Chinese mainland

pixverse/pixverse-v6-r2v

PixVerse V6 reference-to-video model that generates videos from multiple reference images. A comprehensive upgrade over v5.6, recommended for general use cases. Supports one-click conversion of multi-panel storyboard images into videos.PixVerse-Reference-to-Video

Reasoning model

2026-06-01

Chinese mainland

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-05-29

Chinese mainland

vanchin/deepseek-v4-pro

DeepSeek model inference service directly supplied by Kuaishou Wanqing.DeepSeek-Kuaishou Wanqing

Reasoning model

2026-05-29

Chinese mainland

stepfun/step-3.7-flash

Directly supplied by StepFun, Step 3.7 Flash model, the new-generation flagship in the Flash tier. In four key areas: search, Agent, coding, and multimodal are comprehensively upgraded: deep retrieval and multimodal image searchcapability are significantly enhanced, Agent and tool callingcapability are notably improved, codingcapability performs excellently on multiple Flash-tier benchmarks, multimodal understandingcapabilitybenchmarks against top-tier flagship.

Reasoning model

2026-05-25

Chinese mainland

qwen3.7-max-preview, qwen3.7-max-2026-05-17

Models in the Qwen Max series. Text-only input, thinking mode only.

Reasoning model

2026-05-21

Chinese mainland

qwen3.7-max, qwen3.7-max-2026-05-20

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Audio-video translation

2026-05-19

Chinese mainland

qwen3.5-livetranslate-flash-realtime, qwen3.5-livetranslate-flash-realtime-2026-05-19

An audio/video real-time translation model that recognizes 60 source languages and translates audio into 29 languages.Real-time audio and video translation - Qwen

Reasoning model

2026-05-19

Chinese mainland

xiaomi/mimo-v2.5-pro

The MiMo-V2.5-Pro model, directly supplied by Xiaomi, offers substantial enhancements in general agent proficiency, complex software engineering, and long-context task handling. MiMo - Xiaomi

Reasoning model

2026-05-19

Chinese mainland

ZHIPU/GLM-5.1, ZHIPU/GLM-5

The GLM model inference service directly supplied by Zhipu AI. GLM-ZHIPU

Music generation

2026-05-06

Chinese mainland

fun-music-v1

The Fun Music model accepts open-ended song creation requests or lyrics as input to generate full songs with male or female vocals in Chinese or English. Songs are catchy and emotionally rich, combining human inspiration with large model capabilities. Music generation

Reasoning model

2026-04-29

Chinese mainland

deepseek-v4-pro

The unit price of cached_token for deepseek-v4-pro is updated to USD 0.13752131580395 per million tokens. The standard input_token price remains unchanged. See Context Cache.

Reference-to-video

2026-04-29

Chinese mainland

vidu/viduq3-mix_reference2video, vidu/viduq3_reference2video, vidu/viduq3-turbo_reference2video

The Vidu reference-to-video model series supports reference images and text prompts as input, blending the subject from the image into the scene described by the prompt to generate smooth video content. Vidu-reference-to-video

Image-to-video

2026-04-29

Chinese mainland

vidu/viduq2-pro-fast_img2video

The Vidu first-frame video generation model series generates a smooth video based on an input image and text prompts. Vidu: Image-to-video from an initial frame

3D generation

2026-04-29

Chinese mainland

Tripo/Tripo-H3.1

Tripo high-precision 3D generation model. Supports up to 2 million faces for text-to-3D, single image-to-3D, and multi-image-to-3D tasks. Tripo 3D model generation

3D generation

2026-04-29

Chinese mainland

Tripo/Tripo-P1.0

Tripo professional 3D generation model. Supports up to 20,000 faces with fast generation speed for text-to-3D, single image-to-3D, and multi-image-to-3D tasks. Tripo 3D model generation

Text-to-video

2026-04-27

Chinese mainland

happyhorse-1.0-t2v

HappyHorse text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-04-27

Chinese mainland

happyhorse-1.0-i2v

HappyHorse image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-04-27

Chinese mainland

happyhorse-1.0-r2v

HappyHorse reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Video editing

2026-04-27

Chinese mainland

happyhorse-1.0-video-edit

HappyHorse video editing model. Supports video editing and processing. HappyHorse - video editing

Text-to-video

2026-07-01

Chinese mainland

wan2.7-t2v-2026-06-12

A snapshot version of the Wan 2.7 text-to-video model. The model capabilities are the same as wan2.7-t2v. Wan2.7 - text-to-video

Text-to-video

2026-04-26

Chinese mainland

wan2.7-t2v-2026-04-25

A snapshot version of the Wan 2.7 text-to-video model. The model capabilities are the same as wan2.7-t2v. Wan2.7 - text-to-video

Image-to-video

2026-04-26

Chinese mainland

wan2.7-i2v-2026-04-25

A snapshot version of the Wan 2.7 image-to-video model. The model capabilities are the same as wan2.7-i2v. Wan2.7 - image-to-video

Text generation and visual understanding

2026-04-26

Chinese mainland

kimi/kimi-k2.6

kimi/kimi-k2.6 is Kimi's latest and most intelligent model, with comprehensive improvements in general Agent capabilities, coding, and visual understanding. It supports text, image, and video inputs, thinking and non-thinking modes, as well as dialog and Agent tasks.Model inference pricing

Reasoning model

2026-04-24

Chinese mainland

deepseek-v4-pro, deepseek-v4-flash

DeepSeek-V4 series models hosted by Alibaba Cloud (Beijing region). deepseek-v4-pro is a large-scale MoE model with strong general reasoning capabilities. deepseek-v4-flash is a lightweight and cost-effective model optimized for speed.DeepSeek

Image generation

2026-04-23

Chinese mainland

qwen-image-2.0-pro-2026-04-22

A model in the Qwen-Image-2.0 series that unifies image generation and editing. Compared with the March 3 snapshot, this model delivers a significant leap in visual quality — especially texture detail, lighting, and materials. It supports multilingual in-image text generation and delivers more balanced artistic style performance.

Reasoning model

2026-04-23

Chinese mainland

qwen3.5-plus-2026-04-20

A new snapshot of the Qwen3.5 native vision-language Plus model. Compared with the February 15 snapshot, agentic coding capabilities are significantly improved and inference speed is notably faster. Knowledge, reasoning, and long-context capabilities remain at a high level, making it well suited for agentic coding, production workflows, and high-throughput scenarios.

Reasoning model

2026-04-23

Chinese mainland

qwen3.6-27b

A 27B native vision-language dense model in the Qwen3.6 series. Compared with Qwen3.5-27B, agentic coding capabilities are significantly improved, and STEM and reasoning abilities are further enhanced. On the vision side, spatial intelligence and object localization & detection are notably stronger, while video understanding, document OCR, and visual Agent capabilities are steadily improved.

Text generation and visual understanding

2026-04-21

Chinese mainland

kimi-k2.6

The latest and most intelligent model from Moonshot AI, with stronger and more stable long-range code writing capabilities, and significantly improved instruction following and self-correction abilities. Supports text, image, and video input, thinking and non-thinking modes, as well as dialogue and Agent tasks. Kimi - Alibaba Cloud

Reasoning model

2026-04-20

Chinese mainland

qwen3.6-max-preview

The largest proprietary model in the Qwen3.6 series, with further improved coding capabilities and more efficient agent execution. Supports text-only input, thinking mode (enabled by default), explicit caching, and function calling. Overview

Image and video inputs are not supported.

Reasoning model

2026-04-16

Chinese mainland

qwen3.6-flash, qwen3.6-flash-2026-04-16, qwen3.6-35b-a3b

A native vision-language Flash model in the Qwen3.6 series, delivering significant overall performance improvements over Qwen3.5-Flash. Agentic coding capabilities are substantially enhanced (far surpassing the predecessor on multiple code agent benchmarks), along with stronger math and code reasoning. On the vision side, spatial intelligence is notably improved, with object localization and detection being particularly outstanding. Overview

Speech recognition

2026-04-16

Chinese mainland

fun-asr, fun-asr-2025-11-07

Fun-ASR real-time speech recognition capability upgrades:

• Dialect support: Covers the seven major Chinese dialects (Mandarin, Wu, Xiang, Gan, Hakka, Min, Yue) and 20+ regional accents

• Ancient poetry optimization: Improves ancient poetry recognition accuracy for cultural education and audiobook scenarios

• Text optimization: Enhances punctuation prediction and text normalization. Numbers, dates, and amounts automatically convert to standard formats

• Multi-language expansion: Supports 30 languages, including Chinese, English, Japanese, Korean, French, German, and Spanish

For more information, see Audio file recognition - Fun-ASR/Paraformer.

Reasoning model

2026-04-14

Chinese mainland

glm-5.1

The Zhipu GLM-5.1 model is designed for long-horizon tasks and supports a 200K context window with a maximum output of 128K tokens. With strong logical reasoning, long-text comprehension, and code generation capabilities, it delivers excellent results across multiple benchmarks and is well suited for intelligent interaction, enterprise applications, and development assistance. GLM - Alibaba Cloud

Text generation

2026-04-14

Chinese mainland

vanchin/deepseek-v3.2-think, vanchin/deepseek-v3.1-terminus, vanchin/deepseek-r1, vanchin/deepseek-v3

DeepSeek model inference service directly supplied by Kuaishou Wanqing. DeepSeek-Kuaishou Wanqing

OCR

2026-04-14

Chinese mainland

vanchin/deepseek-ocr

DeepSeek OCR model inference service directly supplied by Kuaishou Wanqing. DeepSeek-Kuaishou Wanqing

Text-to-video

2026-04-13

Chinese mainland

pixverse/pixverse-c1-t2v

PixVerse C1 text-to-video model that generates videos from text prompts. Supports intelligent prompt-based storyboarding, enhanced fight scenes and visual effects, and both live-action and anime styles. PixVerse text-to-video

Image-to-video

2026-04-13

Chinese mainland

pixverse/pixverse-c1-it2v

PixVerse C1 image-to-video model that generates videos from an input image and text prompts. Supports intelligent prompt-based storyboarding, enhanced fight scenes and visual effects, and both live-action and anime styles. PixVerse - Image-to-video based on the first frame

Image-to-video

2026-04-13

Chinese mainland

pixverse/pixverse-c1-kf2v

PixVerse C1 keyframe-to-video model that generates smoothly transitioning videos from a start frame image, an end frame image, and text prompts. Supports intelligent prompt-based storyboarding, enhanced fight scenes and visual effects, and both live-action and anime styles. PixVerse: Image-to-video from first and last frames

Reference-to-video

2026-04-13

Chinese mainland

pixverse/pixverse-c1-r2v

PixVerse C1 reference-to-video model that generates videos from multiple reference images. Supports one-click conversion of a multi-panel storyboard collage into a video. PixVerse - Reference-to-video

Video editing

2026-04-03

Chinese mainland

wan2.7-videoedit

The Wan 2.7 video editing model supports instruction-based editing and video migration. It can modify parts of a video or the entire frame, and supports multi-image reference replacement and the replication of actions, effects, and camera movements. Wan2.7 - video editing

Image-to-video

2026-04-03

Chinese mainland

wan2.7-i2v

The Wan 2.7 image-to-video model supports multimodal input, including text, images, audio, and video. It performs three main tasks: generating video from a first frame, generating video from first and last frames, and video continuation. Wan2.7 - image-to-video

Text-to-video

2026-04-03

Chinese mainland

wan2.7-t2v

The Wan 2.7 text-to-video model adds new resolution tiers and custom aspect ratio settings. This allows for flexible adaptation to different creative scenarios and platform publishing needs. Wan2.7 - text-to-video

Reference-to-video

2026-04-03

Chinese mainland

wan2.7-r2v

The Wan 2.7 reference-to-video model supports subject referencing and voice customization. It can generate a scripted video directly from a single, multi-panel storyboard image. Wan2.7 - reference-to-video

Reasoning model

2026-04-02

Chinese mainland

qwen3.6-plus, qwen3.6-plus-2026-04-02

Qwen3.6-Plus features significant upgrades to its code development capabilities, such as Agentic Coding and frontend programming, and provides a significantly enhanced Vibe Coding experience. Its inference capabilities in generalization scenarios are further enhanced. For multimodal capabilities, its performance in universal object recognition, OCR, and object localization is significantly improved. This version also fixes known issues that were identified after Qwen3.5-Plus was published. The usage method is the same as that of qwen3.5-plus. Overview

Image generation and editing

2026-04-01

Chinese mainland

wan2.7-image-pro, wan2.7-image

The Wan 2.7 image generation and editing model supports text-to-image, text-to-image-set, image-to-image-set, image editing, multi-image reference generation, and interactive editing, with improved performance in text rendering, subject consistency, and complex instruction following. The Pro series supports 4K output; the accelerated version balances quality and response speed. Wan - image generation and editing 2.7

Omni-modal

2026-03-30

Chinese mainland

qwen3.5-omni-plus, qwen3.5-omni-plus-2026-03-15, qwen3.5-omni-flash, qwen3.5-omni-flash-2026-03-15

The latest omni-modal model. It supports long video analysis, meeting minutes, subtitle output, security audits, and audio-video interaction. It also supports deep understanding and description generation for audio and video content, recognizes 113 languages, and generates audio in 36 languages. It can process up to 3 hours of audio and 1 hour of video input, and supports web search and instructions to control the volume, speed, and emotion of the output audio. Non-real-time (Qwen-Omni)

Omni-modal

2026-03-30

Chinese mainland

qwen3.5-omni-plus-realtime, qwen3.5-omni-plus-realtime-2026-03-15, qwen3.5-omni-flash-realtime, qwen3.5-omni-flash-realtime-2026-03-15

The latest real-time multimodal model from Qwen. Compared to the previous generation Qwen3-Omni-Flash-Realtime, it features significantly improved intelligence on par with Qwen3.5-Plus. It natively supports web search (WebSearch), voice interruption, and voice control. It recognizes 113 languages and dialects, and generates speech in 36 languages and dialects. Qwen-Omni-Realtime

Text-to-video

2026-03-29

Chinese mainland

pixverse/pixverse-v6-t2v

PixVerse v6 text-to-video model that generates smooth videos from text prompts. PixVerse text-to-video

Image-to-video

2026-03-29

Chinese mainland

pixverse/pixverse-v6-it2v

The PixVerse v6 first-frame-to-video model generates a smooth video based on an input image and a text prompt. PixVerse: Image-to-video based on the first frame

Image-to-video

2026-03-29

Chinese mainland

pixverse/pixverse-v6-kf2v

The PixVerse v6 first-and-last-frame video generation model generates a video with smooth transitions from a first frame, a last frame, and a text prompt. PixVerse - image-to-video - based on first and last frames

Text-to-video

2026-03-27

Chinese mainland

vidu/viduq3-pro_text2video, vidu/viduq3-turbo_text2video, vidu/viduq2_text2video

The Vidu text-to-video model series. These models generate a low-definition video based on text prompts. Vidu text-to-video

Image-to-video

2026-03-27

Chinese mainland

vidu/viduq3-pro_img2video, vidu/viduq3-turbo_img2video, vidu/viduq2-pro_img2video, vidu/viduq2-turbo_img2video

The Vidu first-frame video generation model series generates a smooth video based on an input image and text prompts. Vidu-Image-to-video-based on the first frame

First-and-last-frame-to-video

2026-03-27

Chinese mainland

vidu/viduq3-pro_start-end2video, vidu/viduq3-turbo_start-end2video, vidu/viduq2-pro_start-end2video, vidu/viduq2-turbo_start-end2video

The Vidu start-and-end-frame-to-video model series generates smooth transition videos based on a first frame image, a last frame image, and text prompts. Vidu - Image-to-video - First and last frame

Reference-to-video

2026-03-27

Chinese mainland

vidu/viduq2-pro_reference2video, vidu/viduq2_reference2video

The Vidu reference-to-video model series supports reference images and text prompts as input, blending the subject from the image into the scene described by the prompt to generate smooth video content. Vidu-reference-to-video

Image generation

2026-03-24

Chinese mainland

kling/kling-v3-image-generation, kling/kling-v3-omni-image-generation

The Kling V3 image generation model series supports two types of tasks: text-to-image generation and image-to-image generation. Kling - Image Generation

Video generation

2026-03-24

Chinese mainland

kling/kling-v3-video-generation, kling/kling-v3-omni-video-generation

The Kling V3 video generation model series supports text-to-video, image-to-video based on the first frame, image-to-video based on start and end frames, reference-to-video, and video editing. Kling - Video Generation

Reference-to-video

2026-03-20

Chinese mainland

pixverse/pixverse-v5.6-r2v

Generate a video from multiple reference images. PixVerse - Reference-to-video

Image-to-video

2026-03-20

Chinese mainland

pixverse/pixverse-v5.6-kf2v

Generates a video based on the first and last frames. PixVerse: Image-to-video based on first and last frames

Image-to-video

2026-03-20

Chinese mainland

pixverse/pixverse-v5.6-it2v

Generates video based on the first frame. PixVerse - Image-to-Video - Based on the First Frame

Text-to-video

2026-03-20

Chinese mainland

pixverse/pixverse-v5.6-t2v

A series of text-to-video models. Aishi-Text-to-Video

Multimodal embedding

2026-03-19

Chinese mainland

tongyi-embedding-vision-plus-2026-03-06, tongyi-embedding-vision-flash-2026-03-06

Compared to the main tongyi-embedding-vision model, tongyi-embedding-vision-plus-2026-03-06 / flash-2026-03-06 delivers comprehensive upgrades in visual vectorization. Built on the Qwen3 base, it supports multiple resolutions (0--3 tiers), multiple dimensions (Plus: 64--1152, Flash: 64--768), 30+ languages, and fused text/image/video vectorization. Image search performance improves significantly for IPC, autonomous driving, e-commerce, and video marketing scenarios.

GUI interaction

2026-03-18

Chinese mainland

gui-plus-2026-02-26

Compared to gui-plus, the gui-plus-2026-02-26 model is a comprehensive upgrade that supports thinking and non-thinking modes and delivers significantly improved performance in handling cross-platform and multi-app tasks. Interface interaction

Speech recognition

2026-03-05

Chinese mainland

fun-asr-realtime-2026-02-28

Fun-ASR real-time speech recognition adds a new snapshot model, which performs better than fun-asr-realtime-2025-11-07. Real-time speech recognition - Fun-ASR/Paraformer

Image generation and editing

2026-03-03

Chinese mainland

qwen-image-2.0, qwen-image-2.0-2026-03-03, qwen-image-2.0-pro, qwen-image-2.0-pro-2026-03-03

The Qwen Image 2.0 series supports both image generation and editing. The Pro series offers stronger text rendering, realistic quality, and semantic instruction following. The accelerated version balances quality and response speed. Qwen - text-to-image, Qwen-Image Edit

Speech recognition

2026-03-03

Chinese mainland

qwen3-asr-flash-2026-02-10

Qwen audio file recognition adds a new snapshot model, which performs better than qwen3-asr-flash-2025-09-08. Non-real-time speech recognition

Speech synthesis

2026-03-02

Chinese mainland

cosyvoice-v3.5-plus, cosyvoice-v3.5-flash

The CosyVoice3.5 model is released. It focuses on voice cloning and design, and supports instruction-based control of speech synthesis. Real-time speech synthesis - CosyVoice

Reasoning model

2026-02-24

Chinese mainland

MiniMax-M2.5

This new model from MiniMax responds quickly and excels at tasks such as programming and office work. Usage

Reasoning model

2026-02-24

Chinese mainland

qwen3.5-flash, qwen3.5-flash-2026-02-23, qwen3.5-122b-a10b, qwen3.5-27b, qwen3.5-35b-a3b

The latest Qwen3.5-Flash model and open-source model from Alibaba, supporting text, image, and video input with fast response speed. Its overall performance is close to qwen3.5-plus. Supports built-in Text generation overview. Overview

Code model

2026-02-20

Chinese mainland

qwen3-coder-next

A new-generation open source code generation model in the Qwen3 series. It supports multi-turn tool interactions, and has an enhanced understanding of repository-level code and better adaptability to AI programming tools. Code capabilities (Qwen-Coder)

Reasoning model

2026-02-18

Chinese mainland

glm-5

Zhipu's latest model, designed for programming and agent scenarios, excels at complex systems engineering and long-horizon agent tasks. GLM - Alibaba Cloud

Reasoning model

2026-02-16

Chinese mainland

qwen3.5-plus, qwen3.5-plus-2026-02-15, qwen3.5-397b-a17b

The latest Qwen3.5-Plus and open source models from Alibaba. They support text, image, and video inputs and deliver outstanding performance across multiple tasks, such as language understanding, logical reasoning, code generation, agent tasks, image understanding, video understanding, and graphical user interface (GUI) interaction. They also support built-in tool calling. Overview

Speech recognition

2026-02-13

Chinese mainland

qwen3-asr-flash-realtime-2026-02-10

A new snapshot model for Qwen real-time speech recognition is added, which performs better than qwen3-asr-flash-realtime-2025-10-27. Real-time speech recognition

Speech recognition

2026-02-12

Chinese mainland

fun-asr-flash-8k-realtime, fun-asr-flash-8k-realtime-2026-01-28

A new lightweight ASR model based on the Fun-ASR architecture, optimized for 8 kHz scenarios and suitable for cost-sensitive customers. Real-time speech recognition - Fun-ASR/Paraformer

Speech synthesis

2026-02-10

Chinese mainland

qwen3-tts-instruct-flash, qwen3-tts-instruct-flash-2026-01-26

Qwen speech synthesis introduces an Instruct model, which lets you precisely control the synthesis effect using natural language instructions. Non-real-time speech synthesis

Speech synthesis

2026-02-10

Chinese mainland

qwen3-tts-vd-2026-01-26

Qwen speech synthesis introduces a voice design model, which lets you create custom voices using text descriptions. Non-real-time speech synthesis

Speech synthesis

2026-02-10

Chinese mainland

qwen3-tts-vc-2026-01-22

Qwen speech synthesis introduces a voice cloning model, which lets you quickly clone voices from real audio samples. Non-real-time speech synthesis

Speech synthesis

2026-02-04

Chinese mainland

qwen3-tts-instruct-flash-realtime, qwen3-tts-instruct-flash-realtime-2026-01-22

Qwen real-time speech synthesis adds an Instruct model, which lets you precisely control the synthesis effect using natural language instructions. Real-time speech synthesis

Reference-to-video

2026-02-02

Chinese mainland

wan2.6-r2v-flash

Generates multi-shot videos with automatic dubbing, using a character's appearance from a reference video or image. Wan - Reference-to-Video

Text generation and visual understanding

2026-01-30

Chinese mainland

kimi-k2.5

A visual understanding model from Moonshot AI that excels at general intelligence tasks such as code generation and visual understanding. It supports image, video, and text inputs, and can perform dialogue and agent tasks. Kimi - Alibaba Cloud

Text generation

2026-01-29

Chinese mainland

tongyi-xiaomi-analysis-flash, tongyi-xiaomi-analysis-pro

Tongyi-Xiaomi Dialogue Analysis specializes in analytical tasks such as dialogue information extraction, scenario classification, and satisfaction determination. It excels at handling quality inspection rules with complex business logic, supports custom analysis standards, and has powerful capabilities for multi-turn conversation understanding and semantic inference. Dialogue Analysis (Tongyi-Xiaomi-Analysis)

Speech recognition

2026-01-28

Chinese mainland

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17

The Qwen3-ASR-Flash-Filetrans series models now support word-level timestamps. By setting the new enable_words parameter, you can get millisecond-level word and character alignment information and experience more semantically accurate, fine-grained sentence segmentation. Non-real-time speech recognition

Reasoning model

2026-01-27

Chinese mainland

qwen3-max-2026-01-23

Compared to the snapshot version from September 23, 2025, this model effectively merges thinking and non-thinking modes, significantly enhancing its overall performance. In thinking mode, the model is integrated with three tools: web search, web information extraction, and code interpreter. Using external tools during its thought process, it achieves higher accuracy on complex problems. OpenAI-compatible - Responses

Reasoning model

2026-01-27

Chinese mainland

MiniMax-M2.1

MiniMax-M2.1 is a flagship open source model from MiniMax that focuses on complex real-world tasks. Its core advantages are its multilingual programming capabilities and its ability to solve complex tasks as an Agent. Usage

Visual understanding

2026-01-23

Chinese mainland

qwen3-vl-flash-2026-01-22

A new snapshot of Qwen-VL. Compared to the snapshot from October 15, 2025, it effectively merges thinking and non-thinking modes, significantly enhancing the model's overall performance. It achieves higher inference accuracy in business scenarios such as general visual recognition, security monitoring, store and patrol inspections, and photo-based problem solving. Image and video understanding

Image-to-video

2026-01-17

Chinese mainland

wan2.6-i2v-flash

Supports the generation of videos with and without sound. The two types of videos are billed independently according to their respective billing rules. It also provides multi-shot narrative and audio processing capabilities. Wan - Image-to-video - based on first frame

Image editing

2026-01-17

Chinese mainland

qwen-image-edit-max, qwen-image-edit-max-2026-01-16

The Qwen-Image-Edit-Max series offers more stable and richer editing capabilities, enhanced industrial design and geometric inference capabilities, and improved character consistency and editing precision. Image editing - Qwen

Speech synthesis

2026-01-16

Chinese mainland

qwen3-tts-vc-realtime-2026-01-15

Qwen real-time speech synthesis adds a new snapshot model with further optimized quality. The synthesized voice is more natural and closer to the original compared to qwen3-tts-vc-realtime-2025-11-27. Real-time speech synthesis - Qwen, Real-time speech synthesis

Text-to-image

2026-01-12

Chinese mainland

qwen-image-plus-2026-01-09

This new snapshot model for Qwen text-to-image is a distilled and accelerated version of qwen-image-max that rapidly generates high-quality images. Qwen - text-to-image

Reasoning model

2026-01-12

Chinese mainland

deepseek-v3.2

The deepseek-v3.2 model supports implicit and explicit caching to improve response speed and reduce usage costs without affecting response quality. Context Cache

Speech recognition

2026-01-06

Chinese mainland

qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Qwen3-ASR-Flash supports OpenAI compatible mode. Non-real-time speech recognition

Speech synthesis

2026-01-05

Chinese mainland

cosyvoice-v3-flash

CosyVoice speech synthesis adds 24 new voice timbres (for details, see CosyVoice Voice list):

• Dialects: Long Jiayi, Long Laotie

• Overseas marketing: loongkyong, loongtomoka

• Poetry reading: Long Fei

• Voice assistant: Long Xiaochun, Long Xiaoxia, YUMI

• Social companionship: Long Cheng, Long Ze, Long Zhe, Long Yan, Long Xing, Long Tian, Long Wan, Long Qiang, Long Feifei, Long Hao

• Audiobook: Long Sanshu, Long Yuan, Long Yue, Long Xiu, Long Nan

• News report: Long Shu

Text-to-image

2025-12-31

Chinese mainland

qwen-image-max, qwen-image-max-2025-12-30

The Max series of the Qwen image generation model enhances image realism and naturalness compared to the Plus series, effectively reduces AI-generated artifacts, and excels in areas such as character textures, texture details, and text rendering. Qwen - text-to-image

Reasoning model

2025-12-29

Chinese mainland

glm-4.7

Zhipu AI's latest flagship model features stronger programming capabilities and more stable multi-step reasoning and execution. With 355 billion total parameters, it supports step-by-step planning for complex tasks, code generation, and tool collaboration. It delivers smooth Q\&A, immersive writing, and outstanding creative role-playing performance. GLM - Alibaba Cloud

Image editing

2025-12-23

Chinese mainland

qwen-image-edit-plus-2025-12-15

The latest snapshot model for Qwen image editing enhances character consistency, industrial design capabilities, and geometric inference compared to the previous version. It also optimizes the alignment of spatial layout, texture, and style between the edited and original images, resulting in more precise edits. Image editing - Qwen

Text-to-image

2025-12-19

Chinese mainland

z-image-turbo

A lightweight text-to-image model that quickly generates high-quality images. It supports bilingual rendering in Chinese and English, complex semantic understanding, multiple styles and themes, and flexible adaptation to various resolutions and aspect ratios. Text-to-image Z-Image

Visual understanding

2025-12-19

Chinese mainland

qwen3-vl-plus-2025-12-19

The new Qwen-VL snapshot model features improved instruction-following capabilities and lower latency. Image and video understanding

Speech recognition

2025-12-19

Chinese mainland

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17, qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Speech recognition now supports 9 more languages, including Czech and Danish. Non-real-time speech recognition

Speech recognition

2025-12-17

Chinese mainland

qwen3-asr-flash-realtime, qwen3-asr-flash-realtime-2025-10-27

Speech recognition is now supported in nine additional languages, including Czech and Danish. Real-time speech recognition

Speech recognition

2025-12-17

Chinese mainland

qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Supports audio with any sample rate and sound channel. Non-real-time speech recognition

Speech recognition

2025-12-17

Chinese mainland

fun-asr-mtl, fun-asr-mtl-2025-08-25

Supports speech recognition for 31 languages, including Chinese, English, Japanese, and Korean. Ideal for scenarios in Southeast Asia. Audio file recognition - Fun-ASR/Paraformer

Text-to-image

2025-12-16

Chinese mainland

wan2.6-t2i

A new sync interface is added. It supports selecting custom dimensions within the constraints of total pixel area and aspect ratio. Wan - Text-to-image V2

Image generation and editing

2025-12-16

Chinese mainland

wan2.6-image

Supports image editing and mixed image-text output. Wan - Image Generation and Editing 2.6

Image-to-video - based on first frame

2025-12-16

Chinese mainland

wan2.6-i2v

Adds the multi-shot narrative feature and audio support, allowing you to use automatic dubbing or upload a custom audio file. Wan - Image-to-video - based on first frame

Reference-to-video

2025-12-16

Chinese mainland

wan2.6-r2v

Generates a multi-shot video based on a character's appearance and voice from a reference video. It also supports automatic dubbing. Wan - Reference-to-Video

Text-to-video

2025-12-16

Chinese mainland

wan2.6-t2v

Adds the multi-shot narrative feature, which supports audio using automatic dubbing or custom audio files. Text-to-Video

Voice design

2025-12-16

Chinese mainland

qwen-voice-design

Qwen has released a voice design model for generating customized voices from text descriptions. Use this model with the qwen3-tts-vd-realtime-2025-12-16 model to generate speech in 10 languages. Qwen voice design

Speech synthesis

2025-12-16

Chinese mainland

qwen3-tts-vd-realtime-2025-12-16(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model that supports using Qwen voice design generated voice tones for low-latency, highly stable real-time synthesis; supports multilingual output; automatically adjusts intonation based on text and optimizes synthesis performance for complex texts.Real-time speech synthesis

Speech recognition

2025-12-12

Chinese mainland

fun-asr, fun-asr-2025-11-07

Feature updates for Fun-ASR audio file recognition:

• Supports singing recognition and full song transcription. For details, see Audio file recognition - Fun-ASR/Paraformer.

Speech synthesis

2025-12-11

Chinese mainland

cosyvoice-v3-flash, cosyvoice-v3-plus

• The cosyvoice-v3-flash model now includes five new system voices: longanrou_v3, longyingjing_v3, longyingling_v3, longanling_v3, and longhan_v3. All these voices support the timestamp and SSML features. See CosyVoice Voice list.

• The voice cloning feature for the cosyvoice-v3-flash and cosyvoice-v3-plus models has been enhanced. The feature now supports timestamps and SSML, and provides improved prosody. To try the enhancement, create a new voice. See CosyVoice voice cloning/design API.

Omni-modal

2025-12-04

Chinese mainland

qwen3-omni-flash-2025-12-01

The latest Qwen Omni snapshot model increases the number of supported timbres to 49 and features a significant upgrade to its instruction-following capabilities, enabling it to efficiently understand text, images, audio, and video. Non-real-time (Qwen-Omni)

Real-time multimodal

2025-12-04

Chinese mainland

qwen3-omni-flash-realtime-2025-12-01

The latest snapshot model for the Qwen Omni real-time version offers low-latency multimodal interaction. The number of supported timbres is increased to 49, and the model's instruction-following ability and interactive experience are significantly upgraded. Qwen-Omni-Realtime

Speech translation

2025-12-04

Chinese mainland

qwen3-livetranslate-flash, qwen3-livetranslate-flash-2025-12-01

Qwen3-LiveTranslate-Flash is an audio and video translation model that translates between 18 languages, such as Chinese, English, Russian, and French. It uses visual context to improve translation accuracy and provides both text and speech output. Audio and video file translation - Qwen

Reasoning model

2025-12-04

Chinese mainland

deepseek-v3.2

DeepSeek-V3.2 is the official model that introduces DeepSeek Sparse Attention, a sparse attention mechanism. It is also the first DeepSeek model to integrate thinking with tool usage, and it supports tool calling in both thinking and non-thinking modes.

DeepSeek-Alibaba Cloud

Multilingual translation

2025-11-28

Chinese mainland

qwen-mt-lite

A basic text translation model from Qwen. It supports translation between 31 languages. It offers a faster response time and lower cost than qwen-mt-flash, making it suitable for latency-sensitive scenarios. Machine translation (Qwen-MT)

Voice cloning

2025-11-27

Chinese mainland

qwen-voice-enrollment

Qwen released a voice cloning model. It can quickly generate a highly similar voice from an audio clip of 5 seconds or more. When combined with the qwen3-tts-vc-realtime-2025-11-27 model, it can clone a person's voice with high fidelity and output it in real time across 11 languages. Voice cloning (Qwen)

Speech synthesis

2025-11-27

Chinese mainland

qwen3-tts-vc-realtime-2025-11-27(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model that supports using Qwen voice cloning generated voice tones for low-latency, highly stable real-time synthesis; supports multilingual output; automatically adjusts intonation based on text and optimizes synthesis performance for complex texts.Real-time speech synthesis

Speech synthesis

2025-11-27

Chinese mainland

qwen3-tts-flash-realtime-2025-11-27(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model with low latency and high stability; richer voice tones with multilingual output support for the same voice; automatically adjusts intonation based on text and improves synthesis performance for complex texts.Real-time speech synthesis

Speech synthesis

2025-11-27

Chinese mainland

qwen3-tts-flash-2025-11-27(Snapshot)

Qwen speech synthesis releases a new snapshot model with richer voice tones; the same voice supports multilingual output; adaptively adjusts intonation based on text and optimizes synthesis capability for complex texts.Non-real-time speech synthesis

Text extraction

2025-11-21

Chinese mainland

qwen-vl-ocr-2025-11-20(Snapshot)

Qwen text extraction model. This snapshot version is based on the Qwen3-VL architecture, significantly improving document parsing and text localization capabilities.Text extraction (Qwen-OCR)

Speech recognition

2025-11-20

Chinese mainland

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17(Snapshot)

Qwen audio file recognition releases a new model specifically designed for asynchronous transcription of audio files, supporting recordings up to 12 hours.Non-real-time speech recognition

GUI interaction

2025-11-20

Chinese mainland

gui-plus

GUI-Plus parses user intent from screenshots and natural language instructions, then converts it into standardized graphical user interface (GUI) operations (such as click, input, scroll) for external systems to act upon. It improves GUI operation accuracy compared to Qwen-VL series models. User interface interaction

Speech recognition

2025-11-19

Chinese mainland

fun-asr-realtime-2025-11-07(Snapshot)

Fun-ASR real-time speech recognition releases a new snapshot model, optimizing far-field Voice Activity Detection (VAD) to improve recognition accuracy and stability, and adding support for multiple Chinese dialects and Japanese on top of the existing Chinese and English recognition.Real-time speech recognition - Fun-ASR/Paraformer

Speech recognition

2025-11-19

Chinese mainland

fun-asr-2025-11-07(Snapshot)

Fun-ASR audio file recognition releases a new snapshot model, optimizing far-field Voice Activity Detection (VAD) to improve recognition accuracy and stability, and adding support for multiple Chinese dialects and Japanese on top of the existing Chinese and English recognition.Audio file recognition - Fun-ASR/Paraformer

Speech synthesis

2025-11-19

Chinese mainland

cosyvoice-v3-flash

This version improves pronunciation accuracy and voice similarity compared to previous versions. It also supports more languages, including German, Spanish, French, Italian, and Russian. Real-time speech synthesis - CosyVoice /Sambert

Reasoning model

2025-11-11

Chinese mainland

kimi-k2-thinking

This is a thinking model from Moonshot AI. It has general agent and reasoning capabilities, excels at deep reasoning, and can solve complex problems through multi-step tool calling. Kimi - Alibaba Cloud

Multilingual translation

2025-11-10

Chinese mainland

qwen-mt-flash

Compared to qwen-mt-turbo, this model supports streaming incremental output and has improved overall performance. Machine translation (Qwen-MT)

Reasoning model

2025-11-03

Chinese mainland

qwen3-max-preview

The thinking mode of the qwen3-max-preview model provides significant improvements in overall reasoning capabilities, especially delivering superior performance in agent programming, common-sense reasoning, math, science, and general tasks. Deep thinking

Image editing

2025-10-31

Chinese mainland

qwen-image-edit-plus, qwen-image-edit-plus-2025-10-30

Based on qwen-image-edit, this model offers optimized inference performance and system stability. It greatly reduces the response time for image generation and editing, and supports returning multiple images in a single request. Image editing - Qwen

Real-time speech recognition

2025-10-27

Chinese mainland

qwen3-asr-flash-realtime, qwen3-asr-flash-realtime-2025-10-27

The Qwen real-time speech recognition model features automatic language detection. It can recognize 11 language types and transcribe audio accurately in complex environments. Real-time speech recognition

Visual understanding

2025-10-21

Chinese mainland

qwen3-vl-32b-thinking, qwen3-vl-32b-instruct

A 32B dense model from the Qwen3-VL series. Its overall performance is second only to the Qwen3-VL-235B model. It excels in document recognition and understanding, spatial intelligence, object recognition, 2D visual detection, and spatial reasoning. It is suitable for complex perception tasks in general scenarios. Image and video understanding

Reasoning model

2025-10-21

Chinese mainland

glm-4.6

GLM's next-generation flagship model offers significantly improved core capabilities compared to glm-4.5. It features 355 billion total parameters, 32 billion activated parameters, and a context window extended to 200K. GLM - Alibaba Cloud

Visual understanding

2025-10-16

Chinese mainland

qwen3-vl-flash, qwen3-vl-flash-2025-10-15

A small-sized visual understanding model from the Qwen3 series. It effectively combines thinking and non-thinking modes. Compared to the open source Qwen3-VL-30B-A3B, it performs better and responds faster. Image and video understanding

Visual understanding

2025-10-14

Chinese mainland

qwen3-vl-8b-thinking, qwen3-vl-8b-instruct

An 8B dense open source model from the Qwen3-VL series, available in both thinking and non-thinking versions. It uses less GPU memory and can perform multimodal understanding and reasoning. It supports ultra-long contexts such as long videos and long documents, 2D/3D visual positioning, and comprehensive spatial intelligence and object recognition. Image and video understanding

Visual understanding

2025-10-03

Chinese mainland

qwen3-vl-30b-a3b-thinking, qwen3-vl-30b-a3b-instruct

Based on the new-generation, open source Qwen3-VL model, it is available in both thinking and non-thinking versions. It has a fast response speed and stronger capabilities for multimodal understanding, reasoning, and visual agent tasks. It also supports ultra-long contexts such as long videos and documents. Its spatial intelligence and object recognition capabilities are fully upgraded to handle complex real-world tasks. Image and video understanding

Reasoning model

2025-09-30

Chinese mainland

deepseek-v3.2-exp

A hybrid reasoning model that supports both thinking and non-thinking modes. It introduces a sparse attention mechanism to improve training and inference efficiency for long texts, priced lower than deepseek-v3.1. For details, see DeepSeek-Alibaba Cloud.

Image-to-video

2025-09-23

Chinese mainland

wan2.5-i2v-preview

Added audio capability, supporting automatic dubbing or custom audio files to achieve audio-visual synchronization. Image-to-video - based on the first frame

Text-to-video

2025-09-23

Chinese mainland

wan2.5-t2v-preview

Added audio capabilities for automatic dubbing or importing custom audio files to synchronize audio with video. Text-to-video

Image editing

2025-09-23

Chinese mainland

wan2.5-i2i-preview

Supports inputting text, a single image, or multiple images to achieve entity-consistent image editing, multi-image fusion, and collage generation. General-purpose image editing 2.5

Text-to-image

2025-09-23

Chinese mainland

wan2.5-t2i-preview

Removed single-side restrictions, allowing free dimension selection within the constraints of total pixel area and aspect ratio. Text-to-image V2

Visual understanding

2025-09-23

Chinese mainland

qwen3-vl-plus, qwen3-vl-plus-2025-09-23, qwen3-vl-235b-a22b-thinking, qwen3-vl-235b-a22b-instruct

The Qwen3 series of visual understanding models effectively combines thinking and non-thinking modes, and their visual agent capabilities are world-class. This version features comprehensive upgrades in visual encoding, spatial intelligence, and multimodal thinking. Their visual perception and recognition capabilities are significantly improved. Image and video understanding

Text generation

2025-09-23

Chinese mainland

qwen3-max, qwen3-max-2025-09-23

Compared to the qwen3-max-preview version, this release features specialized upgrades in agent programming and tool calling. The official release achieves state-of-the-art (SOTA) performance in its domain and meets the needs of more complex agent scenarios.

Audio-video translation

2025-09-23

Chinese mainland

qwen3-livetranslate-flash-realtime-2025-09-22

qwen3-livetranslate-flash-realtime is a multilingual real-time audio and video translation model. It can recognize 18 languages and translate them in real time into audio in 10 languages. Real-time audio and video translation - Qwen

Text-to-image

2025-09-23

Chinese mainland

qwen-image-plus

This model excels at rendering complex text, especially Chinese and English. It can create complex mixed-media layouts of images and text. It is more cost-effective than qwen-image. Text-to-image (Qwen-Image)

Code model

2025-09-23

Chinese mainland

qwen3-coder-plus-2025-09-23

Compared to the previous version (July 22 snapshot), this version has improved robustness for downstream tasks and tool calling, and enhanced code security. Code capabilities (Qwen-Coder)

Multimodal embedding

2025-09-23

Chinese mainland

tongyi-embedding-vision-plus, tongyi-embedding-vision-flash

Built on the Qwen series large language models, these models are enhanced for visual vectorization and support three modalities: text, image, and video. Text and multimodal embedding

Speech recognition

2025-09-23

Chinese mainland

fun-asr-realtime, fun-asr-realtime-2025-09-15

Integrates innovative retrieval-augmented generation (RAG) technology and supports multidimensional features such as large-scale custom hotwords, ITN normalization, and punctuation prediction. This significantly improves overall recognition accuracy and contextual relevance. It also supports seamless switching between Chinese and English and has stronger noise robustness, adapting to diverse and complex environments. Real-time speech recognition - Fun-ASR/Paraformer

Real-time speech synthesis

2025-09-22

Chinese mainland

qwen3-tts-flash-realtime, qwen3-tts-flash-realtime-2025-09-18

Qwen's latest real-time speech synthesis large model features 17 highly expressive human-like voices and enables low-latency, high-stability audio synthesis. It also supports multiple languages and dialects. Real-time speech synthesis

Speech synthesis

2025-09-22

Chinese mainland

qwen3-tts-flash, qwen3-tts-flash-2025-09-18

Qwen's latest offline speech synthesis large model features 17 highly expressive human-like voices and enables low-latency, high-stability audio synthesis. It also supports multiple languages and dialects. Non-real-time speech synthesis

Real-time omni-modal

2025-09-22

Chinese mainland

qwen3-omni-flash-realtime, qwen3-omni-flash-realtime-2025-09-15

The real-time version of the Qwen3 series multimodal large model provides low-latency multimodal interaction. It supports streaming audio input and outputs both text and audio, with support for multiple voices and languages. Qwen-Omni-Realtime

Omni-modal

2025-09-22

Chinese mainland

qwen3-omni-flash, qwen3-omni-flash-2025-09-15

A Qwen3 series multimodal model that efficiently understands text, images, audio, and video. It supports text interaction in 119 languages and has excellent instruction-following and system prompt customization capabilities. You can easily set roles and conversation styles to build voice assistants, multimedia analysis, and content creation applications. Non-real-time (Qwen-Omni)

Audio understanding

2025-09-22

Chinese mainland

qwen3-omni-30b-a3b-captioner

An open-source model based on Qwen3-Omni. Without any prompts, it automatically generates precise and comprehensive descriptions for complex audio such as speech, environmental sounds, music, and film and TV sound effects. It can identify speaker emotions, musical elements (such as genre and instruments), and sensitive information, making it suitable for audio content analysis, security review, intent recognition, audio editing, and other fields. Audio understanding (Qwen3-Omni-Captioner)

Image-to-video

2025-09-19

Chinese mainland

wan2.2-animate-move

Transfers the actions and expressions of a character from a template video to a single static character image to generate a character motion video. Wan - Image-to-action

Image-to-video

2025-09-19

Chinese mainland

wan2.2-animate-mix

Replaces the main character in a reference video with the character from an input image, while preserving the original video's scene, lighting, and hue to achieve a seamless character replacement. Wan - video character replacement

Visual understanding

2025-09-12

Chinese mainland

qwen-vl-ocr-2025-08-18

Text localization is fully upgraded, with improved general text recognition and information extraction. Text extraction (Qwen-OCR).

First-and-last-frame-to-video

2025-09-12

Chinese mainland

wan2.2-kf2v-flash

The accuracy of instruction understanding and camera movement control is improved, and overall stability is enhanced. Image-to-video - based on first and last frames

Reasoning model

2025-09-11

Chinese mainland

qwen-plus-2025-09-11

This model is part of the Qwen3 series. Compared to qwen-plus-2025-07-28, it follows instructions better and provides more concise summaries in thinking mode. See Deep thinking. In non-thinking mode, it has enhanced Chinese language understanding and logical reasoning. See Overview.

Reasoning model

2025-09-11

Chinese mainland

qwen3-next-80b-a3b-thinking, qwen3-next-80b-a3b-instruct

A next-generation open-source model based on Qwen3. The thinking modelhas improved instruction following and more concise summaries compared with qwen3-235b-a22b-thinking-2507. See Deep thinking.The instruct modelhas enhanced Chinese comprehension, logical reasoning, and text generation compared with qwen3-235b-a22b-instruct-2507. See Overview.

Speech synthesis

2025-09-08

Chinese mainland

cosyvoice-v3-plus, cosyvoice-v3

CosyVoice speech synthesis adds cosyvoice-v3-plus and cosyvoice-v3 models, which offer better naturalness, sound quality, prosody, and emotional expressiveness compared with previous versions. For usage instructions, see Real-time speech synthesis - CosyVoice /Sambert.

Speech recognition

2025-09-08

Chinese mainland

qwen3-asr-flash(Stable, equivalent to qwen3-asr-flash-2025-09-08)

qwen3-asr-flash-2025-09-08(Snapshot)

Qwen3-ASR model is trained on the Qwen3-Omni base model, supporting multilingual recognition, singing recognition, and noise rejection.Non-real-time speech recognition

Text generation

2025-09-05

Chinese mainland

qwen3-max-preview

The Qwen-Max model (preview), based on Qwen3, offers significant improvements in general capabilities compared to the Qwen 2.5 series. It has significantly enhanced abilities in Chinese and English text understanding, complex instruction following, subjective open-ended tasks, multilingual tasks, and tool calling. The model also has fewer knowledge hallucinations.

Video generation

2025-08-26

Chinese mainland

wan2.2-s2v-detect, wan2.2-s2v

The Wan digital human model supports generating natural-speaking, singing, or performing videos from a single image and audio. It does not restrict character framing and supports portrait, full-body, or half-body character images. Wanxiang - Digital Human

Image translation

2025-08-22

Chinese mainland

qwen-mt-image

The Qwen image translation model supports translating text in images from 11 languages into Chinese or English. It accurately preserves the original layout and content information and provides features such as term definition, sensitive words filter, and image subject detection. Qwen - image translation

Reasoning model

2025-08-22

Chinese mainland

deepseek-v3.1

A hybrid reasoning model that supports both thinking and non-thinking modes. For details, see DeepSeek-Alibaba Cloud.

In-depth research

2025-08-22

Chinese mainland

qwen-deep-research

The Qwen deep research model can break down complex problems, conduct reasoning and analysis using web search, and generate research reports. For more information, see Deep research (Qwen-Deep-Research).

Speech recognition

2025-08-22

Chinese mainland

fun-asr(Stable, equivalent to fun-asr-2025-08-25)

fun-asr-2025-08-25(Snapshot)

Fun-ASR is an end-to-end speech recognition large model developed by Tongyi Lab. Based on advanced proprietary speech technology, it features exceptional context awareness and high-precision transcription capabilities, supporting Chinese and English audio file recognition.Audio file recognition - Fun-ASR/Paraformer

Image editing

2025-08-19

Chinese mainland

qwen-image-edit

The Qwen image editing model supports precise text editing in both Chinese and English, color adjustment, detail enhancement, style transfer, adding or removing objects, and changing positions and actions. It can perform complex image and text editing. Image editing - Qwen

Visual understanding

2025-08-15

Chinese mainland

qwen-vl-plus-2025-08-15

A visual understanding model. It has significantly improved capabilities in object recognition and localization, and multilingual processing. Image and video understanding

Visual understanding

2025-08-13

Chinese mainland

qwen-vl-max-2025-08-13

A visual understanding model. It has comprehensively improved visual understanding metrics and significantly enhanced capabilities in math, reasoning, object recognition, and multilingual processing. Image and video understanding

Text-to-image

2025-08-13

Chinese mainland

qwen-image

The Qwen text-to-image model excels at complex text rendering, especially for Chinese and English text. It can create complex mixed layouts of images and text.Text-to-image (Qwen-Image)

Image-to-video

2025-08-11

Chinese mainland

wan2.2-i2v-flash

Wan 2.2 Flash model. Compared to the 2.1 model, the new version significantly improves image detail and motion stability, with a 50% increase in generation speed. First-frame-to-video

Visual understanding

2025-08-06

Chinese mainland

qwen-vl-plus-2025-07-10

A visual understanding model. Compared to the previous version, it further improves understanding of surveillance video content. Image and video understanding

Code model

2025-08-05

Chinese mainland

qwen3-coder-flash, qwen3-coder-flash-2025-07-28

The fastest and most cost-effective model in the Qwen Coder series. Code capabilities (Qwen-Coder).

Reasoning model

2025-08-05

Chinese mainland

qwen-flash, qwen-flash-2025-07-28

The fastest and most cost-effective model in the Qwen series, suitable for simple jobs.

Reasoning model

2025-07-30

Chinese mainland

qwen-plus-2025-07-28

A model from the Qwen3 series. Compared to the previous version, it increases the context length to 1,000,000. For more information about thinking mode, see Deep thinking. For more information about non-thinking mode, see Overview.

Reasoning model

2025-07-30

Chinese mainland

qwen3-30b-a3b-thinking-2507

qwen3-30b-a3b-instruct-2507

An upgrade of qwen3-30b-a3b. The thinking modelhas improved logical, general, knowledge, and creative capabilities. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Image-to-video

2025-07-28

Chinese mainland

wan2.2-i2v-plus

Compared to the 2.1 model, the new version significantly improves image detail and motion stability. The generation speed increases by up to 50%. First-frame video generation

Text-to-video

2025-07-28

Chinese mainland

wan2.2-t2v-plus

Compared to the 2.1 model, this new version has significantly improved image detail and motion stability. Generation speed is 50% faster. Text-to-video

Text-to-image

2025-07-28

Chinese mainland

wan2.2-t2i-flash, wan2.2-t2i-plus

This new version is a major upgrade to the 2.1 model. It is more creative, stable, and realistic. Images generate 50% faster. Text-to-image

Data mining

2025-07-26

Chinese mainland

qwen-doc-turbo

The Qwen data mining model extracts structured information from documents for data labeling and content moderation. See Data extraction.

Reasoning model

2025-07-24

Chinese mainland

qwen3-235b-a22b-thinking-2507, qwen3-235b-a22b-instruct-2507

An upgrade of qwen3-235b-a22b. The thinking modelhas significantly improved logical, general, knowledge, and creative capabilities, suitable for challenging reasoning scenarios. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Code model

2025-07-23

Chinese mainland

qwen3-coder, qwen3-coder-plus-2025-07-22

A code generation model based on Qwen3. It has strong Coding Agent capabilities and excels at tool calling and environment interaction. It combines excellent code capabilities with general-purpose abilities. Code capabilities (Qwen-Coder)

Speech synthesis

2025-06-26

Chinese mainland

qwen-tts-2025-05-22

The May 22, 2025 snapshot version of the qwen-tts model. Added three new voices: Beijing dialect, Wu dialect, and Sichuan dialect. Non-real-time speech synthesis

Reasoning model

2025-06-24

Chinese mainland

qwen-plus, qwen-turbo

qwen-plus has been updated to support thinking mode, with capabilities identical to qwen-plus-2025-04-28. qwen-turbo has been updated to support thinking mode, with capabilities identical to qwen-turbo-2025-04-28.

Speech synthesis

2025-06-23

Chinese mainland

cosyvoice-v2

cosyvoice-v2 has been fully upgraded with new voices, expanding language support to include languages such as Cantonese, Korean, and Japanese, and introducing more stylized voices such as British English and American English. See CosyVoice Voice list.

Text embedding

2025-06-04

Chinese mainland

text-embedding-v4

text-embedding-v4 is an upgraded model of text-embedding-v3, belonging to the Qwen3-Embedding series. Compared to the previous version, the new model covers more natural languages and multiple programming languages, and adds options for 2048 and 1536 vector dimensions. Text and multimodal embedding

Reasoning model

2025-06-04

Chinese mainland

deepseek-r1-0528

deepseek-r1-0528 is an upgraded version of deepseek-r1. Compared to the previous version, the new model shows significant improvement in complex reasoning tasks. It achieves excellent results in multiple benchmark evaluations for math, programming, and general logic. DeepSeek - Alibaba Cloud Model Studio

Visual reasoning

2025-06-03

Chinese mainland

qvq-max-2025-05-15

A visual reasoning model. Compared to the previous version, it enhances capabilities in math, programming, visual analysis, creation, and general tasks. Visual reasoning

Visual reasoning

2025-06-03

Chinese mainland

qvq-plus, qvq-plus-latest, qvq-plus-2025-05-15

A visual reasoning model. It supports visual input and chain-of-thought output. As the plus version following the qvq-max model, the qvq-plus series offers faster inference speed and a better balance between performance and cost compared to the qvq-max model. Visual reasoning

Speech synthesis

2025-06-03

Chinese mainland

cosyvoice-v2

Speech Synthesis Markup Language (SSML) not only specifies what content to synthesize but also finely controls how it is read, including word and sentence segmentation, pronunciation, speech rate, pauses, intonation, volume, and other vocal features. It even supports adding background music to achieve more natural and expressive speech output. See Introduction to SSML.

Video editing

2025-05-19

Chinese mainland

wanx2.1-vace-plus

A general-purpose video editing model. The model has multimodal input capabilities, combining images, videos, and text prompts. It can perform various tasks such as image-to-video (generating a video based on the subject or background of a reference image) and video repainting (extracting motion features from an input video to generate a new video). General-purpose video editing

Image generation

2025-05-15

Chinese mainland

aitryon-plus

An OutfitAnyone model. Compared to the basic aitryon model, aitryon-plus improves image clarity, clothing texture details, and logo restoration, but takes longer to generate, making it suitable for scenarios where timeliness is not critical. OutfitAnyone - plus edition

Visual understanding

2025-05-15

Chinese mainland

qwen-vl-plus-2025-05-07

A visual understanding model. The model has significantly improved capabilities in math, reasoning, and understanding content from monitoring videos. Image and video understanding

Text extraction

2025-04-30

Chinese mainland

qwen-vl-ocr-2025-04-13

An OCR-specific large model trained on Qwen-VL that provides image and text recognition capabilities. This model adds six built-in OCR tasks: general recognition, document parsing, table parsing, information extraction, multilingual recognition, and formula recognition. Text extraction

Reasoning model

2025-04-28

Chinese mainland

Qwen3 commercial models

qwen-plus-2025-04-28, qwen-turbo-2025-04-28

Qwen3Open-source models

qwen3-235b-a22b, qwen3-30b-a3b, qwen3-32b, qwen3-14b, qwen3-8b, qwen3-4b, qwen3-1.7b, qwen3-0.6b

Qwen3 models support thinking mode and non-thinking mode. You can use the enable_thinking parameter to switch between the two modes. In addition, the capabilities of Qwen3 models have been significantly enhanced:

1. Reasoning capability: In math, code, and logical reasoning benchmarks, it significantly surpasses QwQ and non-reasoning models of the same size, reaching the industry-leading level at the same scale.

2. Human preference capability: Creative writing, role-playing, multi-turn conversation, and instruction-following capabilities are all significantly improved, with general capabilities notably surpassing models of the same size.

3. Agent capability: Reaches industry-leading level in both reasoning and non-reasoning modes, capable of precisely calling external tools.

4. Multilingual capability: Supports over 100 languages and dialects, with significant improvements in multilingual translation, instruction understanding, and commonsense reasoning.

5. Response format fix: Fixed response format issues in previous versions, such as abnormal Markdown, mid-response truncation, and incorrect boxed output.

For thinking mode, seeDeep thinking. For non-thinking mode, seeOverview.

Voice cloning

2025-04-24

Chinese mainland

cosyvoice-v2

Users only need to provide 10--20 seconds of audio to quickly generate a highly similar and naturally sounding custom voice. Voice cloning/design

Visual understanding

2025-04-18

Chinese mainland

qwen-vl-max-2025-04-08

A visual understanding model. Math and reasoning abilities are improved, the response style is adjusted to align with human preferences, and the detail and format clarity of model responses are significantly improved. Image and video understanding

Image-to-video

2025-04-18

Chinese mainland

wanx2.1-kf2v-plus

Based on the input first and last frame images, the model can generate a smooth and fluid dynamic video according to the prompt. First-and-last-frame-to-video

Visual understanding

2025-04-04

Chinese mainland

qwen-vl-max-2025-04-02

A visual understanding model. Accuracy in solving complex math problems is significantly improved. The response style is greatly adjusted to align with human preferences, especially for objective questions such as math, logical reasoning, and knowledge Q\&A. The detail and format clarity of the model's responses are also significantly improved. Image and video understanding

Visual reasoning

2025-03-28

Chinese mainland

qvq-max, qvq-max-latest, qvq-max-2025-03-25

A visual reasoning model. It supports visual input and chain-of-thought output, demonstrating stronger capabilities in math, programming, visual analysis, creation, and general tasks. Visual reasoning

Omni-modal

2025-03-26

Chinese mainland

qwen-omni-turbo-2025-03-26

A new multimodal understanding and generation model from Qwen. It supports text, image, speech, and video input, and outputs text and audio. It provides four natural conversational voices. See Non-real-time (Qwen-Omni).

Omni-modal

2025-03-26

Chinese mainland

qwen2.5-omni-7b

A new multimodal model from Qwen for understanding and generation. It supports text, image, speech, and video input, and outputs text and audio. It provides two natural conversational voices. Non-real-time (Qwen-Omni).

Image editing

2025-03-25

Chinese mainland

wanx2.1-imageedit

A general-purpose image editing model. It uses single-sentence instructions to provide various image editing features, such as image outpainting, text watermark removal, image inpainting, and style transfer. Wan - General-purpose Image Editing 2.1

Visual understanding

2025-03-24

Chinese mainland

qwen2.5-vl-32b-instruct 

Visual understanding model. Achieves near Qwen2.5VL-72B level in mathematical problem solving. Response style is significantly adjusted for human preferences, especially for objective questions such as math, logical reasoning, and knowledge QA. Response detail and format clarity are significantly improved.Image and video understanding

Reasoning model

2025-03-06

Chinese mainland

qwq-plus, qwq-plus-latest, qwq-plus-2025-03-05

A QwQ reasoning model trained on the Qwen2.5 model. It greatly improves model inference capability through reinforcement learning. The model's core metrics for math and code (AIME 24/25, LiveCodeBench) and some general metrics (IFEval, LiveBench, etc.) reach the level of the full-performance DeepSeek-R1. Deep thinking

Open-source reasoning model

2025-03-06

Chinese mainland

qwq-32b

A QwQ reasoning model trained on the Qwen2.5-32B model. It greatly improves the model's reasoning ability through reinforcement learning. The model's core metrics for math and code (AIME 24/25, LiveCodeBench) and some general metrics (IFEval, LiveBench, etc.) reach the level of the full-performance DeepSeek-R1. All metrics significantly exceed those of DeepSeek-R1-Distill-Qwen-32B, which is also based on Qwen2.5-32B. Deep thinking

Speech recognition/translation

2025-03-03

Chinese mainland

gummy-realtime-v1

gummy-chat-v1

The Gummy large model supports real-time speech recognition and translation, accurately recognizing 10 languages, such as Chinese, English, Japanese, and Korean. Additionally, it supports mutual translation between Chinese, English, Japanese, and Korean, and unidirectional translation of six other languages into Chinese or English. See Real-time speech recognition - Fun-ASR/Paraformer.

Image-to-video

2025-02-25

Chinese mainland

wanx2.1-i2v-turbo

Compared with the wanx2.1-i2v-plus model, the turbo model generates faster (one-third the time of the plus model) at a better cost-performance ratio. For usage instructions, see Wan - image-to-video - first frame API reference (2.1-2.6).

Omni-modal

2025-02-14

Chinese mainland

qwen-omni-turbo

qwen-omni-turbo-latest

qwen-omni-turbo-2025-01-19

The Qwen-Omni series models support input of multiple modalities, including video, audio, images, and text, and output text. See Non-real-time (Qwen-Omni).

Visual understanding

2025-02-07

Chinese mainland

qwen-vl-max-2025-01-25

A visual understanding model belonging to the Qwen2.5-VL model series. Compared to the previous version, it extends the context to 128k and significantly enhances image and video understanding capabilities.

Visual understanding

2025-02-07

Chinese mainland

qwen-vl-plus-2025-01-25

A visual understanding model belonging to the Qwen2.5-VL series. Compared to the previous version, it extends the context to 128k, significantly enhancing image and video understanding capabilities.

Text generation

2025-01-27

Chinese mainland

deepseek-v3

deepseek-r1

DeepSeek series models are large language models launched by DeepSeek.

• DeepSeek-V3 is an MoE model with 671B parameters and 37B activated parameters. It was pretrained on 14.8T tokens and excels in long text, code, math, encyclopedia, and Chinese capabilities.

• DeepSeek-R1 extensively uses reinforcement learning in the post-training phase. With minimal labeled data, it greatly enhances model reasoning capabilities, especially in math, code, and natural language reasoning tasks.

• See DeepSeek - Alibaba Cloud Model Studio.

Visual understanding

2025-01-27

Chinese mainland

qwen2.5-vl-3b-instruct

qwen2.5-vl-7b-instruct

qwen2.5-vl-72b-instruct

• Compared to the Qwen2-VL model, this model has the following improvements:

  • It has significantly improved capabilities in instruction following, mathematical calculations, code generation, and structured output (JSON output).

  • It supports unified parsing of visual content in images, such as text, charts, and layouts. The model can also accurately locate visual elements and represent them using detection frames and coordinate points.

  • It understands long video files (up to 10 minutes), provides event localization with second-level precision, and recognizes the sequence and speed of events.

• See Image and video understanding.

Text generation

2025-01-27

Chinese mainland

qwen-max-2025-01-25

qwen2.5-14b-instruct-1m

qwen2.5-7b-instruct-1m

• The qwen-max-2025-01-25 model (also known as Qwen2.5-Max): The best-performing model in the Qwen series. It has significant improvements in code writing and comprehension, logical reasoning, and multilingual capabilities. The response style is greatly adjusted to align with human preferences. The detail and format clarity of its responses are significantly improved, with targeted enhancements in content creation, JSON format adherence, and role assumption. See Overview of text generation models.

• The qwen2.5-14b-instruct-1m and qwen2.5-7b-instruct-1m models: Compared to the qwen2.5-14b-instruct and qwen2.5-7b-instruct models, the context length is increased to 1,000,000. See Overview of text generation models.

Image-to-video

2025-01-22

Chinese mainland

emoji-detect-v1

emoji-v1

• Generates dynamic face videos based on face images and preset face motion templates. This model can be used for emoji creation, video material generation, and other scenarios. See Emoji quickstart.

Text generation

2025-01-17

Chinese mainland

qwen-plus-2025-01-12

• Compared to the qwen-plus-2024-12-20 model, this model has enhanced overall capabilities in both Chinese and English, with significant improvements in common sense and reading comprehension. Its ability to switch naturally between different languages, dialects, and styles is significantly improved, and Chinese instruction following is significantly enhanced. See Qwen.

Image-to-video

2025-01-17

Chinese mainland

wanx2.1-i2v-plus

• Takes an image as the first frame and generates a video based on prompts. For usage instructions, see Wan - image-to-video - first frame API reference (2.1-2.6).

Text-to-image

2025-01-17

Chinese mainland

wanx2.0-t2i-turbo

• Excels at realistic portraits and creative design, with medium speed and high cost-effectiveness. See Text-to-Image V2 series models.

Visual understanding

2025-01-13

Chinese mainland

qwen-vl-plus-2025-01-02

• Compared to the qwen-vl-plus-0809 model, it significantly improves instruction following, image understanding, and math capabilities. See Image and video understanding.

Text-to-video

2025-01-08

Chinese mainland

wanx2.1-t2v-turbo

wanx2.1-t2v-plus

• Generate a video from a single sentence.

• It has strong instruction-following capabilities, supports large and complex movements, and replicates real-world physics. The generated videos feature rich artistic styles and cinematic-quality visuals. See Wanx - text-to-video.

Text-to-image

2025-01-08

Chinese mainland

wanx2.1-t2i-turbo

wanx2.1-t2i-plus

• Text-to-Image V2 series models are fully upgraded Wan text-to-image models. We recommend trying them.

Visual understanding

2025-01-07

Chinese mainland

qwen-vl-max-2024-12-30

• This model enriches its knowledge base, further improves image recognition and understanding, and can parse complex visual content. For usage instructions, see Image and video understanding.

Speech recognition

2025-01-02

Chinese mainland

qwen-audio-asr

qwen-audio-asr-latest

qwen-audio-asr-2024-12-04

• Qwen ASR is a speech recognition model trained on Qwen-Audio, Currently supported languages: Chinese and English. For usage instructions, see Real-time speech recognition - Fun-ASR/Paraformer.

Visual reasoning

2024-12-25

Chinese mainland

qvq-72b-preview

• Focused on enhancing visual reasoning capabilities, especially in the realm of mathematical reasoning. See Image and video understanding.

Multilingual translation

2024-12-25

Chinese mainland

qwen-mt-plus

qwen-mt-turbo

• The Qwen-MT model is a machine translation large language model optimized based on the Qwen model. It excels at translation between Chinese and English, Chinese and minor languages, and English and minor languages. The minor languages include 26 languages such as Japanese, Korean, French, Spanish, German, Portuguese (Brazil), Thai, Indonesian, Vietnamese, and Arabic. In addition to multilingual translation, the model supports features such as terminology intervention, domain prompting, and translation memory to improve translation quality in complex scenarios. See Machine translation (Qwen-MT).

Visual understanding

2024-12-18

Chinese mainland

qwen2-vl-72b-instruct

• Achieved state-of-the-art results on multiple visual understanding benchmarks, significantly enhancing the processing of multimodal tasks. See Image and video understanding for instructions.

Intention recognition

2024-12-12

Chinese mainland

tongyi-intent-detect-v3

• The Qwen intent recognition model can quickly and accurately parse user intent within hundreds of milliseconds and select the appropriate tool to solve user problems. See Intent recognition.

Vocal avatar

2024-12-10

Chinese mainland

videoretalk

• Supports generating lip-sync videos based on character videos and audio. For details.

Dancing Portrait

2024-12-10

Chinese mainland

animate-anyone-gen2

animate-anyone-detect-gen2

animate-anyone-template-gen2

• Provides character image compliance detection, character motion template generation, and character video generation. Calling these three models in sequence can generate character dance videos. For details, see AnimateAnyone: Generate dance videos from images.

Speech synthesis

2024-12-10

Chinese mainland

cosyvoice-v1

• Users only need to provide 10-20 seconds of audio to quickly generate a custom voice that is highly similar and natural-sounding. For details, see Voice cloning/design.

Llama series

2024-12-09

Chinese mainland

llama3.3-70b-instruct

• Added the 70B parameter model from the third-party Llama 3.3 series.

Audio understanding

2024-12-09

Chinese mainland

qwen-audio-turbo-latest

qwen-audio-turbo-2024-12-04

• The new model significantly improves speech recognition accuracy compared to the qwen-audio-turbo-2024-08-07 version and adds voice chat capabilities. See Audio understanding - Qwen-Audio.

Text generation

2024-11-28

Chinese mainland

qwq-32b-preview

• This model focuses on enhancing AI reasoning capabilities. SeeQwQ.

2024-11-28

Chinese mainland

qwen-plus-2024-11-25

qwen-plus-2024-11-27

• Compared to the qwen-plus-0919 model, Chinese and English response detail is significantly improved, better aligning with user preferences; role-playing capability is significantly enhanced; Chinese text creation capability is significantly improved; Chinese and English instruction-following capability is improved. Fixed RAG citation superscript generation issues in certain scenarios. For more information, seeModel Studio console.

2024-11-15

Chinese mainland

qwen-turbo-2024-11-01

• Context length extended to one million tokens. For more information, seeModel Studio console.

Visual understanding

2024-11-15

Chinese mainland

qwen-vl-max-2024-11-19

• Improved image understanding and reduced repetitive text generation. For details, see Image and video understanding.

Chinese mainland

qwen-vl-max-2024-10-30

• Enhanced multilingual understanding capabilities.For more information, seeImage and video understanding.

Chinese mainland

• qwen-vl-ocr

• qwen-vl-ocr-2024-10-28

• qwen-vl-ocr-latest

• qwen-vl-ocr is a model dedicated to OCR; text extraction capabilities for table, exam, and similar image types are significantly improved.For more information, seeText extraction (Qwen-OCR).

Text generation

2024-11-12

Chinese mainland

• Proprietary models such as qwen-coder-plus

• Open-source models such as qwen2.5-coder-32b-instruct

• These models demonstrate industry-leading capabilities in code generation, code repair, and code reasoning.For more information, seeCode capabilities (Qwen-Coder).

LivePortrait

2024-11-7

Chinese mainland

• liveportrait-detect

• liveportrait

• Generate dynamic portrait videos from portrait images and voice audio files in a fast and lightweight manner. For more information, see LivePortrait quick start.

Vibrant Portrait

2024-11-7

Chinese mainland

• emo-detect-v1

• emo-v1

• Generates a dynamic portrait video from a portrait image and a voice audio file. This service can be invoked directly without requiring deployment. It differs from the previous EMO models (emo-detect and emo) in its invocation and billing methods. See EMO quickstart.

Llama series

2024-11-5

Chinese mainland

• llama3.2-90b-vision-instruct

• llama3.2-11b-vision

• Added third-party Large Language Models (LLMs) from the Llama 3.2 series with 11B and 90B parameters. These models include the visual understanding feature.

Visual understanding

2024-10-29

Chinese mainland

qwen2-vl-2b-instruct

• Extended the context to 32k to significantly improve image understanding capabilities. See Image and video understanding.

Visual understanding

2024-10-23

Chinese mainland

qwen2-vl-7b-instruct

• Added the qwen2-vl-7b-instruct model. See Image and video understanding.

Text generation

2024-09-18

Chinese mainland

qwen-long

• Added the qwen-long model. For more information, seeLong context (Qwen-Long)

Visual understanding

2024-08-27

Chinese mainland

qwen-vl-max-0809

• This model is the August 9, 2024 snapshot version of qwen-vl-max. It will be updated to the main qwen-vl-max version on September 9. The snapshot version will be maintained for one month after the next snapshot version is released (to be determined). For complete information, see Qwen-VL API Details.

Text generation

2024-08-16

Chinese mainland

Qwen2-Math series models

• Has strong math problem-solving capabilities.

Text-to-image

2024-08-07

Chinese mainland

FLUX text-to-image model

• The FLUX text-to-image model is a high-quality open-source text-to-image generation model from Black Forest Labs. It demonstrates exceptional performance in multiple dimensions, especially in text-guided image generation, multi-subject scene construction, and detailed hand generation, setting a new technical benchmark for the text-to-image field.

Llama series

2024-07-23

Chinese mainland

• Llama3.1-8b-instruct

• Llama3.1-70b-instruct

• Llama3.1-405b-instruct

• Added Llama 3.1 series model versions.

Text embedding

2024-07-10

Chinese mainland

text-embedding-v3

• The text-embedding-v3 model is an upgrade of the text-embedding-v2 model, featuring high performance, low cost, 50+ language support, and extra-long text.

Main updates include the following:

• Language expansion : Compared to text-embedding-v2 , text-embedding-v3 adds support for languages such as Italian, Polish, Vietnamese, and Thai, increasing the total number of supported languages to over 50.

• Input length extension : The supported input length for encoding is extended from 2048 to 8192. For text-embedding-v3, 8192 refers to the maximum token length of the input text.

• Variable output continuous vector dimensions : Unlike text-embedding-v2, which has a fixed 1536-dimensional vector, text-embedding-v3 allows users to customize the continuous vector dimension, with options for 512, 768, and 1024 dimensions. Additionally, to further reduce downstream task costs, text-embedding-v3 lowers the maximum vector dimension to 1024 without degrading performance.

• No distinction between Query/Document types : text-embedding-v3 no longer distinguishes between input text types without reducing model effectiveness, so the text_type parameter no longer needs to specify whether the input text is a Query or Document type.

• Sparse vector support : text-embedding-v3 supports both continuous vector representation (dense vector) and discrete vector representation (sparse vector). Users can specify in the interface parameters to output continuous vectors, discrete vectors, or both.

• Performance improvement : The performance of the embedding model has been improved by optimizing the pre-trained model base and the SFT policy, as demonstrated by evaluations on public datasets.

Text generation

2024-07-05

Chinese mainland

• qwen-turbo-0206

• qwen-turbo-0624

• Added snapshot versions of the qwen-turbo model. See Text generation.

Text generation

2024-07-05

Chinese mainland

• qwen-plus-0206

• qwen-plus-0624

• Added snapshot versions for the qwen-plus model. See Text generation.

Text generation

2024-07-02

Chinese mainland

Minimax large language model

• Added the Minimax large language model.

Text generation

2024-06-07

Chinese mainland

qwen2 series open-source large language models

• Compared to Qwen1.5, Qwen2 outperforms most open source models in benchmarks such as language understanding, language generation, multilingual capabilities, coding, and inference.

Text generation

2024-06-03

Chinese mainland

The Yi large language model

The Yi large language model is a large language model with hundreds of billions of parameters. It is the only Chinese large model in the top 10 of the LM SYS leaderboard. It has powerful Q&A, inference, and text generation capabilities.

Text generation

2024-04-28

Chinese mainland

qwen1.5-110b-chat

• qwen1.5-110b-chat is a 110-billion-parameter open-source chat model from Qwen1.5 that is aligned with human instructions. See Qwen open source models.

Type

Time

Deployment scope

Model

Description

Reference-to-video

2026-07-01

International

wan2.7-r2v-2026-06-12

A snapshot version of the Wan 2.7 reference-to-video model. It supports subject referencing and voice customization, and can generate a scripted video directly from a single, multi-panel storyboard image. Wan2.7 - reference-to-video

Reasoning model

2026-06-26

International

kimi-k2.7-code

Kimi K2.7 Code model is now available in the Singapore region. An agent-centric coding model optimized for long-range software engineering tasks. Supports thinking mode only. Kimi - Alibaba Cloud

Image generation

2026-06-25

International

qwen-image-2.0-pro-2026-06-22

The latest snapshot of the Qwen-Image-2.0 series, unifying image generation and editing. Compared with the April 22 snapshot, this model features enhanced text rendering with support for up to 1k token instructions, more refined photorealistic quality and scene detail, and stronger semantic adherence.

Text-to-video

2026-06-22

International

happyhorse-1.1-t2v

HappyHorse 1.1 text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-06-22

International

happyhorse-1.1-i2v

HappyHorse 1.1 image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-06-22

International

happyhorse-1.1-r2v

HappyHorse 1.1 reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Reasoning model

2026-06-10

International

qwen3.7-max-2026-06-08

The Max model, the largest and most capable in the Qwen3.7 series, has added visual-modal understanding compared to the May 20 snapshot, enabling it to perceive real-world scenes and supporting multimodal interactive hybrid agent capabilities.

Reasoning model

2026-06-01

International

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-05-27

International

glm-5.1

The Zhipu GLM-5.1 model is designed for long-horizon tasks and supports a 200K context window with a maximum output of 128K tokens. With strong logical reasoning, long-text comprehension, and code generation capabilities, it delivers excellent results across multiple benchmarks and is well suited for intelligent interaction, enterprise applications, and development assistance. GLM - Alibaba Cloud

Reasoning model

2026-05-21

International

qwen3.7-max, qwen3.7-max-2026-05-20

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Audio-video translation

2026-05-19

International

qwen3.5-livetranslate-flash-realtime, qwen3.5-livetranslate-flash-realtime-2026-05-19

An audio/video real-time translation model that recognizes 60 source languages and translates audio into 29 languages.Real-time audio and video translation - Qwen

Text-to-video

2026-04-27

International

happyhorse-1.0-t2v

HappyHorse text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-04-27

International

happyhorse-1.0-i2v

HappyHorse image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-04-27

International

happyhorse-1.0-r2v

HappyHorse reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Video editing

2026-04-27

International

happyhorse-1.0-video-edit

HappyHorse video editing model. Supports video editing and processing. HappyHorse - video editing

Text-to-video

2026-07-01

International

wan2.7-t2v-2026-06-12

A snapshot version of the Wan 2.7 text-to-video model. The model capabilities are the same as wan2.7-t2v. Wan2.7 - text-to-video

Text-to-video

2026-04-26

International

wan2.7-t2v-2026-04-25

A snapshot version of the Wan 2.7 text-to-video model. The model capabilities are the same as wan2.7-t2v. Wan2.7 - text-to-video

Image-to-video

2026-04-26

International

wan2.7-i2v-2026-04-25

A snapshot version of the Wan 2.7 image-to-video model. The model capabilities are the same as wan2.7-i2v. Wan2.7 - image-to-video

Image generation

2026-04-23

International

qwen-image-2.0-pro-2026-04-22

A model in the Qwen-Image-2.0 series that unifies image generation and editing. Compared with the March 3 snapshot, this model delivers a significant leap in visual quality — especially texture detail, lighting, and materials. It supports multilingual in-image text generation and delivers more balanced artistic style performance.

Reasoning model

2026-04-23

International

qwen3.5-plus-2026-04-20

A new snapshot of the Qwen3.5 native vision-language Plus model. Compared with the February 15 snapshot, agentic coding capabilities are significantly improved and inference speed is notably faster. Knowledge, reasoning, and long-context capabilities remain at a high level, making it well suited for agentic coding, production workflows, and high-throughput scenarios.

Reasoning model

2026-04-23

International

qwen3.6-27b

A 27B native vision-language dense model in the Qwen3.6 series. Compared with Qwen3.5-27B, agentic coding capabilities are significantly improved, and STEM and reasoning abilities are further enhanced. On the vision side, spatial intelligence and object localization & detection are notably stronger, while video understanding, document OCR, and visual Agent capabilities are steadily improved.

Reasoning model

2026-04-20

International

qwen3.6-max-preview

The largest proprietary model in the Qwen3.6 series, with further improved coding capabilities and more efficient agent execution. Supports text-only input, thinking mode (enabled by default), explicit caching, and function calling. Overview

Image and video inputs are not supported.

Reasoning model

2026-04-16

International

qwen3.6-flash, qwen3.6-flash-2026-04-16, qwen3.6-35b-a3b

A native vision-language Flash model in the Qwen3.6 series, delivering significant overall performance improvements over Qwen3.5-Flash. Agentic coding capabilities are substantially enhanced (far surpassing the predecessor on multiple code agent benchmarks), along with stronger math and code reasoning. On the vision side, spatial intelligence is notably improved, with object localization and detection being particularly outstanding. Overview

Speech recognition

2026-04-16

International

fun-asr, fun-asr-2025-11-07

Fun-ASR real-time speech recognition capability upgrades:

• Dialect support: Covers the seven major Chinese dialects (Mandarin, Wu, Xiang, Gan, Hakka, Min, Yue) and 20+ regional accents

• Ancient poetry optimization: Improves ancient poetry recognition accuracy for cultural education and audiobook scenarios

• Text optimization: Enhances punctuation prediction and text normalization. Numbers, dates, and amounts automatically convert to standard formats

• Multi-language expansion: Supports 30 languages, including Chinese, English, Japanese, Korean, French, German, and Spanish

For more information, see Audio file recognition - Fun-ASR/Paraformer.

Video editing

2026-04-03

International

wan2.7-videoedit

The Wan 2.7 video editing model supports instruction-based editing and video migration. It can modify parts of a video or the entire frame, and supports multi-image reference replacement and the replication of actions, effects, and camera movements. Wan2.7 - video editing

Image-to-video

2026-04-03

International

wan2.7-i2v

The Wan 2.7 image-to-video model supports multimodal input, including text, images, audio, and video. It performs three main tasks: generating video from a first frame, generating video from first and last frames, and video continuation. Wan2.7 - image-to-video

Text-to-video

2026-04-03

International

wan2.7-t2v

The Wan 2.7 text-to-video model adds new resolution tiers and custom aspect ratio settings. This allows for flexible adaptation to different creative scenarios and platform publishing needs. Wan2.7 - text-to-video

Reference-to-video

2026-04-03

International

wan2.7-r2v

The Wan 2.7 reference-to-video model supports subject referencing and voice customization. It can generate a scripted video directly from a single, multi-panel storyboard image. Wan2.7 - reference-to-video

Reasoning model

2026-04-02

International

qwen3.6-plus, qwen3.6-plus-2026-04-02

Qwen3.6-Plus features significant upgrades to its code development capabilities, such as Agentic Coding and frontend programming, and provides a significantly enhanced Vibe Coding experience. Its inference capabilities in generalization scenarios are further enhanced. For multimodal capabilities, its performance in universal object recognition, OCR, and object localization is significantly improved. This version also fixes known issues that were identified after Qwen3.5-Plus was published. The usage method is the same as that of qwen3.5-plus. Overview

Image generation and editing

2026-04-01

International

wan2.7-image-pro, wan2.7-image

The Wan 2.7 image generation and editing model supports text-to-image, text-to-image-set, image-to-image-set, image editing, multi-image reference generation, and interactive editing, with improved performance in text rendering, subject consistency, and complex instruction following. The Pro series supports 4K output; the accelerated version balances quality and response speed. Wan - image generation and editing 2.7

Omni-modal

2026-03-30

International

qwen3.5-omni-plus, qwen3.5-omni-plus-2026-03-15, qwen3.5-omni-flash, qwen3.5-omni-flash-2026-03-15

The latest omni-modal model. It supports long video analysis, meeting minutes, subtitle output, security audits, and audio-video interaction. It also supports deep understanding and description generation for audio and video content, recognizes 113 languages, and generates audio in 36 languages. It can process up to 3 hours of audio and 1 hour of video input, and supports web search and instructions to control the volume, speed, and emotion of the output audio. Non-real-time (Qwen-Omni)

Omni-modal

2026-03-30

International

qwen3.5-omni-plus-realtime, qwen3.5-omni-plus-realtime-2026-03-15, qwen3.5-omni-flash-realtime, qwen3.5-omni-flash-realtime-2026-03-15

The latest real-time multimodal model from Qwen. Compared to the previous generation Qwen3-Omni-Flash-Realtime, it features significantly improved intelligence on par with Qwen3.5-Plus. It natively supports web search (WebSearch), voice interruption, and voice control. It recognizes 113 languages and dialects, and generates speech in 36 languages and dialects. Qwen-Omni-Realtime

Reasoning model

2026-03-20

International

deepseek-v3.2

DeepSeek-V3.2 is the official model that introduces DeepSeek Sparse Attention, a sparse attention mechanism. It is the first model from DeepSeek that integrates thinking with tool calling. The model supports tool calling in both thinking and non-thinking modes.

DeepSeek-Alibaba Cloud

Image generation and editing

2026-03-03

International

qwen-image-2.0, qwen-image-2.0-2026-03-03, qwen-image-2.0-pro, qwen-image-2.0-pro-2026-03-03

The Qwen Image 2.0 series supports both image generation and editing. The Pro series offers stronger text rendering, realistic quality, and semantic instruction following. The accelerated version balances quality and response speed. Qwen - text-to-image, Qwen-Image Edit

Speech recognition

2026-03-03

International

qwen3-asr-flash-2026-02-10

Qwen audio file recognition adds a new snapshot model, which performs better than qwen3-asr-flash-2025-09-08. Non-real-time speech recognition

Reasoning model

2026-02-24

International

qwen3.5-flash, qwen3.5-flash-2026-02-23, qwen3.5-122b-a10b, qwen3.5-27b, qwen3.5-35b-a3b

The latest Qwen3.5-Flash model and open-source model from Alibaba, supporting text, image, and video input with fast response speed. Its overall performance is close to qwen3.5-plus. Supports built-in Text generation overview. Overview

Code model

2026-02-20

International

qwen3-coder-next

A new-generation open source code generation model in the Qwen3 series. It supports multi-turn tool interactions, and has an enhanced understanding of repository-level code and better adaptability to AI programming tools. Code capabilities (Qwen-Coder)

Reasoning model

2026-02-16

International

qwen3.5-plus, qwen3.5-plus-2026-02-15, qwen3.5-397b-a17b

The latest Qwen3.5-Plus and open source models from Alibaba. They support text, image, and video inputs and deliver outstanding performance across multiple tasks, such as language understanding, logical reasoning, code generation, agent tasks, image understanding, video understanding, and graphical user interface (GUI) interaction. They also support built-in tool calling. Overview

Speech recognition

2026-02-13

International

qwen3-asr-flash-realtime-2026-02-10

A new snapshot model for Qwen real-time speech recognition is added, which performs better than qwen3-asr-flash-realtime-2025-10-27. Real-time speech recognition

Speech synthesis

2026-02-10

International

cosyvoice-v3-plus, cosyvoice-v3-flash

CosyVoice speech synthesis adds v3 models that support speech synthesis with system voices and cloned voices. Real-time speech synthesis - CosyVoice /Sambert

Speech synthesis

2026-02-10

International

qwen3-tts-instruct-flash, qwen3-tts-instruct-flash-2026-01-26

Qwen speech synthesis introduces an Instruct model, which lets you precisely control the synthesis effect using natural language instructions. Non-real-time speech synthesis

Speech synthesis

2026-02-10

International

qwen3-tts-vd-2026-01-26

Qwen speech synthesis introduces a voice design model, which lets you create custom voices using text descriptions. Non-real-time speech synthesis

Speech synthesis

2026-02-10

International

qwen3-tts-vc-2026-01-22

Qwen speech synthesis introduces a voice cloning model, which lets you quickly clone voices from real audio samples. Non-real-time speech synthesis

Speech synthesis

2026-02-04

International

qwen3-tts-instruct-flash-realtime, qwen3-tts-instruct-flash-realtime-2026-01-22

Qwen real-time speech synthesis adds an Instruct model, which lets you precisely control the synthesis effect using natural language instructions. Real-time speech synthesis

Reference-to-video

2026-02-02

International

wan2.6-r2v-flash

Generates multi-shot videos with automatic dubbing, using a character's appearance from a reference video or image. Wan - Reference-to-Video

Visual understanding

2026-01-28

International

qwen3-vl-flash-2026-01-22

This is a new snapshot of the Qwen-VL model. It merges thinking and non-thinking modes to improve overall performance compared to the October 15, 2025 snapshot. The model achieves higher inference accuracy in scenarios such as general visual recognition, security, store patrols, inspections, and solving problems from photos. Image and video understanding

Speech recognition

2026-01-28

International

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17

The Qwen3-ASR-Flash-Filetrans series models now support word-level timestamps. By setting the new enable_words parameter, you can get millisecond-level word and character alignment information and experience more semantically accurate, fine-grained sentence segmentation. Non-real-time speech recognition

Reasoning model

2026-01-27

International

qwen3-max-2026-01-23

Compared to the snapshot version from September 23, 2025, this model effectively merges thinking and non-thinking modes, significantly enhancing its overall performance. In thinking mode, the model is integrated with three tools: web search, web information extraction, and code interpreter. Using external tools during its thought process, it achieves higher accuracy on complex problems. OpenAI-compatible - Responses

Image-to-video

2026-01-18

International

wan2.6-i2v-flash

Generates videos with or without audio. The two video types are billed independently based on their respective billing rules. It also has multi-shot narrative and audio processing capabilities. Wan - Image-to-video - Based on first frame

Image editing

2026-01-18

International

qwen-image-edit-max, qwen-image-edit-max-2026-01-16

The Qwen Image Edit Max series provides more stable and versatile editing capabilities, enhances industrial design and geometric inference, and improves character consistency and editing precision. Image editing - Qwen

Speech synthesis

2026-01-16

International

qwen3-tts-vc-realtime-2026-01-15

Qwen real-time speech synthesis adds a new snapshot model with further optimized quality. The synthesized voice is more natural and closer to the original compared to qwen3-tts-vc-realtime-2025-11-27. Real-time speech synthesis - Qwen, Real-time speech synthesis

Text-to-image

2026-01-12

International

qwen-image-plus-2026-01-09

This new snapshot model for Qwen text-to-image is a distilled and accelerated version of qwen-image-max that rapidly generates high-quality images. Qwen - text-to-image

Speech recognition

2026-01-06

International

qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Qwen3-ASR-Flash supports OpenAI compatible mode. Non-real-time speech recognition

Text-to-image

2025-12-31

International

qwen-image-max, qwen-image-max-2025-12-30

The Max series of the Qwen image generation model enhances image realism and naturalness compared to the Plus series, effectively reduces AI-generated artifacts, and excels in areas such as character textures, texture details, and text rendering. Qwen - text-to-image

Image editing

2025-12-23

International

qwen-image-edit-plus-2025-12-15

The latest snapshot model for Qwen image editing enhances character consistency, industrial design capabilities, and geometric inference compared to the previous version. It also optimizes the alignment of spatial layout, texture, and style between the edited and original images, resulting in more precise edits. Image editing - Qwen

Text-to-image

2025-12-22

International

z-image-turbo

A lightweight text-to-image model that quickly generates high-quality images. It supports bilingual rendering in Chinese and English, complex semantic understanding, multiple styles and themes, and flexibly adapts to various resolutions and aspect ratios. Z-Image

Visual understanding

2025-12-19

International

qwen3-vl-plus-2025-12-19

The new Qwen-VL snapshot model features improved instruction-following capabilities and lower latency. Image and video understanding

Speech recognition

2025-12-19

International

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17, qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Speech recognition now supports 9 more languages, including Czech and Danish. Non-real-time speech recognition

Speech recognition

2025-12-17

International

qwen3-asr-flash-realtime, qwen3-asr-flash-realtime-2025-10-27

Speech recognition is now supported in nine additional languages, including Czech and Danish. Real-time speech recognition

Speech recognition

2025-12-17

International

qwen3-asr-flash, qwen3-asr-flash-2025-09-08

Supports audio with any sample rate and sound channel. Non-real-time speech recognition

Speech recognition

2025-12-17

International

fun-asr-mtl, fun-asr-mtl-2025-08-25

Supports speech recognition for 31 languages, including Chinese, English, Japanese, and Korean. Ideal for scenarios in Southeast Asia. Audio file recognition - Fun-ASR/Paraformer

Voice design

2025-12-16

International

qwen-voice-design

Qwen has released a voice design model for generating customized voices from text descriptions. Use this model with the qwen3-tts-vd-realtime-2025-12-16 model to generate speech in 10 languages. Qwen voice design

Speech synthesis

2025-12-16

International

qwen3-tts-vd-realtime-2025-12-16(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model that supports using Qwen voice design generated voice tones for low-latency, highly stable real-time synthesis; supports multilingual output; automatically adjusts intonation based on text and optimizes synthesis performance for complex texts.Real-time speech synthesis

Text-to-image

2025-12-16

International

wan2.6-t2i

A new sync interface is added. It supports selecting custom dimensions within the constraints of total pixel area and aspect ratio. Wan - Text-to-image V2

Image generation and editing

2025-12-16

International

wan2.6-image

Supports image editing and mixed image-text output. Wan - Image Generation and Editing 2.6

Image-to-video - based on first frame

2025-12-16

International

wan2.6-i2v

Adds the multi-shot narrative feature and audio support, allowing you to use automatic dubbing or upload a custom audio file. Wan - Image-to-video - based on first frame

Reference-to-video

2025-12-16

International

wan2.6-r2v

Generates a multi-shot video based on a character's appearance and voice from a reference video. It also supports automatic dubbing. Wan - Reference-to-Video

Text-to-video

2025-12-16

International

wan2.6-t2v

Adds the multi-shot narrative feature, which supports audio using automatic dubbing or custom audio files. Text-to-Video

Speech recognition

2025-12-12

International

fun-asr, fun-asr-2025-11-07

Feature updates for Fun-ASR audio file recognition:

• Supports singing recognition and full song transcription. For details, see Audio file recognition - Fun-ASR/Paraformer.

Omni-modal

2025-12-04

International

qwen3-omni-flash-2025-12-01

The latest Qwen Omni snapshot model increases the number of supported timbres to 49 and features a significant upgrade to its instruction-following capabilities, enabling it to efficiently understand text, images, audio, and video. Non-real-time (Qwen-Omni)

Real-time multimodal

2025-12-04

International

qwen3-omni-flash-realtime-2025-12-01

The latest snapshot model for the Qwen Omni real-time version offers low-latency multimodal interaction. The number of supported timbres is increased to 49, and the model's instruction-following ability and interactive experience are significantly upgraded. Qwen-Omni-Realtime

Speech translation

2025-12-04

International

qwen3-livetranslate-flash, qwen3-livetranslate-flash-2025-12-01

Qwen3-LiveTranslate-Flash is an audio and video translation model that translates between 18 languages, such as Chinese, English, Russian, and French. It uses visual context to improve translation accuracy and provides both text and speech output. Audio and video file translation - Qwen

Multilingual translation

2025-12-02

International

qwen-mt-lite

The Qwen basic text translation model supports translation between 31 languages. Compared to qwen-mt-flash, this model provides faster responses at a lower cost, making it suitable for latency-sensitive scenarios. Machine translation (Qwen-MT)

Voice cloning

2025-11-27

International

qwen-voice-enrollment

Qwen released a voice cloning model. It can quickly generate a highly similar voice from an audio clip of 5 seconds or more. When combined with the qwen3-tts-vc-realtime-2025-11-27 model, it can clone a person's voice with high fidelity and output it in real time across 11 languages. Voice cloning (Qwen)

Speech synthesis

2025-11-27

International

qwen3-tts-vc-realtime-2025-11-27(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model that supports using Qwen voice cloning generated voice tones for low-latency, highly stable real-time synthesis; supports multilingual output; automatically adjusts intonation based on text and optimizes synthesis performance for complex texts.Real-time speech synthesis

Speech synthesis

2025-11-27

International

qwen3-tts-flash-realtime-2025-11-27(Snapshot)

Qwen real-time speech synthesis releases a new snapshot model with low latency and high stability; richer voice tones with multilingual output support for the same voice; automatically adjusts intonation based on text and improves synthesis performance for complex texts.Real-time speech synthesis

Speech synthesis

2025-11-27

International

qwen3-tts-flash-2025-11-27(Snapshot)

Qwen speech synthesis releases a new snapshot model with richer voice tones; the same voice supports multilingual output; adaptively adjusts intonation based on text and optimizes synthesis capability for complex texts.Non-real-time speech synthesis

Text extraction

2025-11-21

International

qwen-vl-ocr-2025-11-20(Snapshot)

Qwen text extraction model. This snapshot version is based on the Qwen3-VL architecture, significantly improving document parsing and text localization capabilities.Text extraction (Qwen-OCR)

Speech recognition

2025-11-20

International

qwen3-asr-flash-filetrans, qwen3-asr-flash-filetrans-2025-11-17(Snapshot)

Qwen audio file recognition releases a new model specifically designed for asynchronous transcription of audio files, supporting recordings up to 12 hours.Non-real-time speech recognition

Speech recognition

2025-11-19

International

fun-asr-2025-11-07(Snapshot)

Fun-ASR audio file recognition releases a new snapshot model, optimizing far-field Voice Activity Detection (VAD) to improve recognition accuracy and stability, and adding support for multiple Chinese dialects and Japanese on top of the existing Chinese and English recognition.Audio file recognition - Fun-ASR/Paraformer

Multilingual translation

2025-11-11

International

qwen-mt-flash

Compared to qwen-mt-turbo, this model supports streaming incremental output and offers improved overall performance. Machine translation (Qwen-MT)

Image-to-video

2025-11-10

International

wan2.2-animate-move

Transfers the actions and expressions of a character from a template video to a single static image to generate a video of the character in motion. Wan - Image-to-action

Image-to-video

2025-11-10

International

wan2.2-animate-mix

This model replaces the main character in a reference video with a character from an image. It preserves the original video's scene, lighting, and tone for seamless character replacement. Wan - Video Character Swap

Reasoning model

2025-11-03

International

qwen3-max-preview

The thinking mode of the qwen3-max-preview model provides significant improvements in overall reasoning capabilities, especially delivering superior performance in agent programming, common-sense reasoning, math, science, and general tasks. Deep thinking

Image editing

2025-10-31

International

qwen-image-edit-plus, qwen-image-edit-plus-2025-10-30

Based on qwen-image-edit, this model offers optimized inference performance and system stability. It greatly reduces the response time for image generation and editing, and supports returning multiple images in a single request. Image editing - Qwen

Reasoning model

2025-10-31

International

qwen3-next-80b-a3b-thinking, qwen3-next-80b-a3b-instruct

A next-generation open-source model based on Qwen3. The thinking modelhas improved instruction following and more concise summaries compared with qwen3-235b-a22b-thinking-2507. See Deep thinking.The instruct modelhas enhanced Chinese comprehension, logical reasoning, and text generation compared with qwen3-235b-a22b-instruct-2507. See Overview.

Real-time speech recognition

2025-10-27

International

qwen3-asr-flash-realtime, qwen3-asr-flash-realtime-2025-10-27

The Qwen real-time speech recognition model features automatic language detection. It can recognize 11 language types and transcribe audio accurately in complex environments. Real-time speech recognition

Text, image, video, speech, etc.

2025-09-23

International

Model list

First launch in the Singapore region.

Type

Time

Deployment scope

Model

Description

Reasoning model

2026-07-06

US

qwen3.7-plus-us

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-07-03

US

qwen3.7-max-us

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Text-to-video

2026-06-26

Global

happyhorse-1.1-t2v

HappyHorse 1.1 text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-06-26

Global

happyhorse-1.1-i2v

HappyHorse 1.1 image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-06-26

Global

happyhorse-1.1-r2v

HappyHorse 1.1 reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Reasoning model

2026-06-10

Global

qwen3.7-max-2026-06-08

The Max model, the largest and most capable in the Qwen3.7 series, has added visual-modal understanding compared to the May 20 snapshot, enabling it to perceive real-world scenes and supporting multimodal interactive hybrid agent capabilities.

Reasoning model

2026-06-02

Global

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-05-21

Global

qwen3.7-max, qwen3.7-max-2026-05-20

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Reasoning model

2026-05-20

Global

glm-5.1

Zhipu GLM-5.1 model, designed for long-range tasks with 200K context support and up to 128K token output. With powerful logical reasoning, long-text understanding, and code generation capabilities, it achieves excellent results on multiple benchmarks, suitable for intelligent interaction, enterprise applications, and development assistance scenarios.GLM - Alibaba Cloud

Text-to-video

2026-05-06

Global

happyhorse-1.0-t2v

HappyHorse text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-05-06

Global

happyhorse-1.0-i2v

HappyHorse image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Video editing

2026-05-06

Global

happyhorse-1.0-video-edit

HappyHorse video editing model. Supports video editing and processing. HappyHorse - video editing

Reference-to-video

2026-05-06

Global

happyhorse-1.0-r2v

HappyHorse reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Text generation and visual understanding

2026-04-29

Global

kimi-k2.5

A visual understanding model from Moonshot AI that excels in general-purpose intelligent tasks such as code generation and visual understanding. It also supports image, video, and text input, along with dialogue and Agent tasks. Kimi - Alibaba Cloud

Reasoning model

2026-04-29

Global

deepseek-v4-pro

The unit price of cached_token for deepseek-v4-pro is updated to USD 0.13752131580395 per million tokens. The standard input_token price remains unchanged. See Context Cache.

Multilingual translation

2026-04-10

US

qwen-mt-lite-us

A basic-tier Qwen text translation model supporting mutual translation across 31 languages. Compared with qwen-mt-flash, it provides faster responses at lower cost, suitable for latency-sensitive scenarios. Machine translation (Qwen-MT)

Reasoning model

2026-04-02

Global

qwen3.6-plus, qwen3.6-plus-2026-04-02

Qwen3.6-Plus features significant upgrades to its code development capabilities, such as Agentic Coding and frontend programming, and provides a significantly enhanced Vibe Coding experience. Its inference capabilities in generalization scenarios are further enhanced. For multimodal capabilities, its performance in universal object recognition, OCR, and object localization is significantly improved. This version also fixes known issues that were identified after Qwen3.5-Plus was published. The usage method is the same as that of qwen3.5-plus. Overview

Reasoning model

2026-03-04

Global

qwen3.5-flash, qwen3.5-flash-2026-02-23, qwen3.5-122b-a10b, qwen3.5-27b, qwen3.5-35b-a3b

The latest Qwen3.5-Flash model and open-source model from Alibaba, supporting text, image, and video input with fast response speed. Its overall performance is close to qwen3.5-plus. Supports built-in Text generation overview. Overview

Reasoning model

2026-03-04

Global

qwen3.5-plus, qwen3.5-plus-2026-02-15, qwen3.5-397b-a17b

The latest Qwen3.5-Plus model and open-source model from Alibaba, supporting text, image, and video input. It delivers outstanding performance across language understanding, logical reasoning, code generation, agent tasks, image understanding, video understanding, GUI, and other tasks. Supports built-in Text generation overview. Overview

Image-to-video - based on first frame

2026-01-12

US

wan2.6-i2v-us

Adds the multi-shot narrative capability and supports audio with automatic dubbing or the use of custom audio files. Wan - Image-to-video based on the first frame

Text-to-video

2026-01-12

US

wan2.6-t2v-us

Added multi-shot narrative capability. Supports audio capabilities, including automatic dubbing or custom audio file input. Wan2.7 - text-to-video

Speech recognition

2026-01-12

US

qwen3-asr-flash-us, qwen3-asr-flash-2025-09-08-us

Supports audio with any sample rate and sound channel. Non-real-time speech recognition

Visual understanding

2026-01-12

US

qwen3-vl-flash-us, qwen3-vl-flash-2025-10-15-us

A small-sized visual understanding model from the Qwen3 series that effectively fuses thinking and non-thinking modes. It provides better performance and faster responses than the open source Qwen3-VL-30B-A3B version. Image and video understanding

Reasoning model

2026-01-12

US

qwen-plus-2025-12-01-us

This model is part of the Qwen3 series. Compared to qwen-plus-2025-07-28, it has improved instruction-following capabilities and provides more concise summaries in thinking mode. See Deep thinking. In non-thinking mode, it has enhanced Chinese language understanding and logical reasoning capabilities. See Overview.

Reasoning model

2026-01-12

US

qwen-plus-us

Offers a balance of inference performance, cost, and speed between Qwen-Max and Qwen-Flash, making it suitable for tasks of medium complexity.

Text generation

2026-01-12

US

qwen-flash-us, qwen-flash-2025-07-28-us

The fastest and most affordable model in the Qwen series. It is suitable for simple tasks.

Text-to-image

2026-01-12

Global

wan2.6-t2i

Adds a synchronous API. Supports custom dimensions within total pixel area and aspect ratio constraints. Wan - Text-to-Image V2

Image generation and editing

2026-01-12

Global

wan2.6-image

Supports image editing and mixed image-text output. Wan 2.6: Image generation and editing

Image-to-video - based on first frame

2026-01-12

Global

wan2.6-i2v

Adds multi-shot narrative capability. Supports audio, including automatic dubbing or using custom audio files. Wan image-to-video - based on first frame

Reference-to-video

2026-01-12

Global

wan2.6-r2v

Generates multi-shot videos based on the character and voice from a reference video. Supports automatic dubbing. Wan reference-to-video

Text-to-video

2026-01-12

Global

wan2.6-t2v

Adds multi-shot narrative capability. Supports audio, including automatic dubbing or using custom audio files. Wan text-to-video

Visual understanding

2026-01-12

Global

qwen3-vl-flash, qwen3-vl-flash-2025-10-15

A small-sized visual understanding model from the Qwen3 series that effectively combines thinking and non-thinking modes. It delivers better performance and faster response times than the open source Qwen3-VL-30B-A3B model. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-30b-a3b-thinking, qwen3-vl-30b-a3b-instruct

Based on the new-generation open source Qwen3-VL model, this model is available in thinking and non-thinking versions. It features fast response times and enhanced capabilities for multimodal understanding, inference, visual agent tasks, and extra-long context support for long videos and documents. Its fully upgraded spatial intelligence and object recognition make it suitable for complex real-world tasks. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-8b-thinking, qwen3-vl-8b-instruct

This is an 8B dense open-source model from the Qwen3-VL series, available in thinking and non-thinking versions. It uses less GPU memory and performs multimodal understanding and inference. It supports extra-long context for long videos and documents, 2D/3D visual positioning, and comprehensive spatial intelligence and object recognition. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-32b-thinking, qwen3-vl-32b-instruct

A 32B dense model from the Qwen3-VL series. Its overall performance is second only to the Qwen3-VL-235B model. It excels in document recognition and understanding, spatial intelligence, object recognition, 2D visual detection, and spatial reasoning. It is suitable for complex perception tasks in general scenarios. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-plus, qwen3-vl-plus-2025-09-23, qwen3-vl-235b-a22b-thinking, qwen3-vl-235b-a22b-instruct

The Qwen3 series visual understanding model effectively combines thinking and non-thinking modes. Its visual agent capabilities are world-class. This version features comprehensive upgrades in visual encoding, spatial intelligence, and multimodal thinking. Its visual perception and recognition capabilities are significantly improved. Image and video understanding

Reasoning model

2026-01-12

Global

qwen3-next-80b-a3b-thinking, qwen3-next-80b-a3b-instruct

A next-generation open-source model based on Qwen3. The thinking modelhas improved instruction following and more concise summaries compared with qwen3-235b-a22b-thinking-2507. See Deep thinking.The instruct modelhas enhanced Chinese comprehension, logical reasoning, and text generation compared with qwen3-235b-a22b-instruct-2507. See Overview.

Reasoning model

2026-01-12

Global

qwen3-max, qwen3-max-2025-09-23

Compared to the qwen3-max-preview version, this release features specialized upgrades in agent programming and tool calling. The official release achieves state-of-the-art (SOTA) performance in its domain and meets the needs of more complex agent scenarios.

Reasoning model

2026-01-12

Global

qwen3-max-preview

The Qwen-Max preview model, built on Qwen3, delivers a substantial improvement in general capabilities over the Qwen2.5 series. It shows significant gains in Chinese and English text understanding, complex instruction following, subjective open-ended tasks, multilingual tasks, and tool calling. It also exhibits fewer knowledge hallucinations. Qwen-Max

Code model

2026-01-12

Global

qwen3-coder-flash, qwen3-coder-flash-2025-07-28

The fastest and most cost-effective model in the Qwen Coder series. Code capabilities (Qwen-Coder).

Code model

2026-01-12

Global

qwen3-coder-plus-2025-09-23

Compared to the previous version (July 22 snapshot), this model has improved robustness for downstream tasks and tool calling, along with enhanced code security. Code generation (Qwen-Coder)

Code model

2026-01-12

Global

qwen3-coder-plus, qwen3-coder-plus-2025-07-22, qwen3-coder-30b-a3b-instruct, qwen3-coder-480b-a35b-instruct

A code generation model based on Qwen3. It has strong Coding Agent capabilities, excels at tool calling and environment interaction, and combines excellent coding skills with general-purpose abilities. Code generation (Qwen-Coder)

Reasoning model

2026-01-12

Global

qwen3-30b-a3b-thinking-2507, qwen3-30b-a3b-instruct-2507

An upgrade of qwen3-30b-a3b. The thinking modelhas improved logical, general, knowledge, and creative capabilities. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Reasoning model

2026-01-12

Global

qwen3-235b-a22b-thinking-2507, qwen3-235b-a22b-instruct-2507

An upgrade of qwen3-235b-a22b. The thinking modelhas significantly improved logical, general, knowledge, and creative capabilities, suitable for challenging reasoning scenarios. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Reasoning model

2026-01-12

Global

qwen3-235b-a22b, qwen3-30b-a3b, qwen3-32b, qwen3-14b, qwen3-8b

Qwen3 models support both thinking and non-thinking modes. You can switch between the two modes by using the enable_thinking parameter. In addition, the capabilities of Qwen3 models are significantly improved:

1. 1. Reasoning ability: In evaluations for math, code, and logical reasoning, it significantly outperforms QwQ and other non-reasoning models of the same size, reaching top-tier industry performance for its scale.

2. 2. Human preference alignment: Creative writing, role-playing, multi-turn conversation, and instruction-following capabilities are greatly improved. Its general capabilities significantly exceed those of other models of the same size.

3. 3. Agent capabilities: Achieves industry-leading performance in both reasoning and non-reasoning modes and can accurately invoke external tools.

4. 4. Multilingual capabilities: Supports over 100 languages and dialects, with notable improvements in multilingual translation, instruction understanding, and common-sense reasoning.

5. 5. Response format fixes: Fixes response format issues from previous versions, such as abnormal Markdown, mid-response truncation, and incorrect boxed output.

For thinking mode, see Deep thinking. For non-thinking mode, see Overview.

Text extraction

2026-01-12

Global

qwen-vl-ocr-2025-11-20

The Qwen text extraction model. This snapshot version is based on the Qwen3-VL architecture and significantly improves document parsing and text localization capabilities. Text extraction

Text extraction

2026-01-12

Global

qwen-vl-ocr

qwen-vl-ocr is a model specialized for OCR, with significantly improved text extraction from tables, exam papers, and similar images. For details, see Text extraction (Qwen-OCR).

Reasoning model

2026-01-12

Global

qwen-plus-2025-12-01, qwen-plus-2025-09-11

This model is part of the Qwen3 series. Compared to qwen-plus-2025-07-28, it provides enhanced instruction-following capabilities and more concise summaries and responses in thinking mode. See Deep thinking. In non-thinking mode, its Chinese language understanding and logical reasoning capabilities are enhanced. See Overview.

Reasoning model

2026-01-12

Global

qwen-plus-2025-07-28

This model is part of the Qwen3 series. Compared to the previous version, its context length is increased to 1,000,000. For thinking mode, see Deep thinking. For non-thinking mode, see Overview.

Reasoning model

2026-01-12

Global

qwen-plus

Offers a balance of capabilities. Its inference performance, cost, and speed are between those of Qwen-Max and Qwen-Flash, making it suitable for moderately complex tasks.

Multilingual translation

2026-01-12

Global

qwen-mt-lite

A basic-level model from Qwen for text translation. It supports translation between 31 languages. Compared to qwen-mt-flash, it has faster response times and lower costs, making it suitable for latency-sensitive scenarios. Machine translation (Qwen-MT)

Multilingual translation

2026-01-12

Global

qwen-mt-plus, qwen-mt-flash

The Qwen-MT model is a machine translation large language model that is optimized based on the Qwen model. It excels at translation between Chinese and English, between Chinese and other languages, and between English and other languages. The 26 other languages include Japanese, Korean, French, Spanish, German, Portuguese (Brazil), Thai, Indonesian, Vietnamese, and Arabic. In addition to multilingual translation, it provides features such as terminology intervention, domain prompts, and translation memory to improve translation quality in complex application scenarios. See Machine translation (Qwen-MT).

Text generation

2026-01-12

Global

qwen-flash, qwen-flash-2025-07-28

The fastest and most cost-effective model in the Qwen series, suitable for simple jobs.

Type

Time

Deployment scope

Model

Description

Text-to-video

2026-06-26

Global

happyhorse-1.1-t2v

HappyHorse 1.1 text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-06-26

Global

happyhorse-1.1-i2v

HappyHorse 1.1 image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Reference-to-video

2026-06-26

Global

happyhorse-1.1-r2v

HappyHorse 1.1 reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Reasoning model

2026-06-10

Global

qwen3.7-max-2026-06-08

The Max model, the largest and most capable in the Qwen3.7 series, has added visual-modal understanding compared to the May 20 snapshot, enabling it to perceive real-world scenes and supporting multimodal interactive hybrid agent capabilities.

Reasoning model

2026-06-02

Global

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-05-21

Global

qwen3.7-max, qwen3.7-max-2026-05-20

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Reasoning model

2026-05-20

Global

glm-5.1

Zhipu GLM-5.1 model, designed for long-range tasks with 200K context support and up to 128K token output. With powerful logical reasoning, long-text understanding, and code generation capabilities, it achieves excellent results on multiple benchmarks, suitable for intelligent interaction, enterprise applications, and development assistance scenarios.GLM - Alibaba Cloud

Reasoning model

2026-05-07

EU

qwen3-max, qwen3-max-2026-01-23

Compared to the September 23, 2025 snapshot version, it effectively integrates thinking mode and non-thinking mode, significantly improving the overall model performance.

Reasoning model

2026-05-07

EU

qwen-plus, qwen-plus-2025-12-01

Belongs to the Qwen3 series. Compared to qwen-plus-2025-07-28, it improves instruction-following capability in thinking mode with more concise summary responses. For details, seeDeep thinking.Chinese understanding and logical reasoning capabilities are enhanced in non-thinking mode. For details, seeOverview.

Reasoning model

2026-05-07

EU

qwen3.5-flash, qwen3.5-flash-2026-02-23

The latest Alibaba model Qwen3.5-Flash and open-source model, supporting text, image, and video input with fast response speed.Overview

Visual understanding

2026-05-07

EU

qwen3-vl-plus, qwen3-vl-flash, qwen3-vl-flash-2025-10-15

Qwen3 series visual understanding model that effectively integrates thinking mode and non-thinking mode, achieving world-leading visual agent capabilities. This version features comprehensive upgrades in visual encoding, spatial perception, and multimodal thinking; visual perception and recognition capabilities are significantly improved, with stronger instruction-following capability and lower latency.Image and video understanding

Code capability

2026-05-07

EU

qwen3-coder-next

Qwen3 series new generation open-source code generation model, supporting multi-turn tool interaction, with improved understanding of repository-level code and adaptability to AI programming tools.Code capabilities (Qwen-Coder)

Text-to-video

2026-05-06

Global

happyhorse-1.0-t2v

HappyHorse text-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - text-to-video

Image-to-video

2026-05-06

Global

happyhorse-1.0-i2v

HappyHorse image-to-video model. Supports audio video generation with 3-15 seconds duration at 720P/1080P resolution. HappyHorse image-to-video (first frame)

Video editing

2026-05-06

Global

happyhorse-1.0-video-edit

HappyHorse video editing model. Supports video editing and processing. HappyHorse - video editing

Reference-to-video

2026-05-06

Global

happyhorse-1.0-r2v

HappyHorse reference-to-video model. Supports multiple reference images as input to generate audio videos with 3-15 seconds duration at 720P/1080P resolution. HappyHorse - reference-to-video

Text generation and visual understanding

2026-04-29

Global

kimi-k2.5

A visual understanding model from Moonshot AI that excels in general-purpose intelligent tasks such as code generation and visual understanding. It also supports image, video, and text input, along with dialogue and Agent tasks. Kimi - Alibaba Cloud

Reasoning model

2026-04-29

Global

deepseek-v4-pro

The unit price of cached_token for deepseek-v4-pro is updated to USD 0.13752131580395 per million tokens. The standard input_token price remains unchanged. See Context Cache.

Reasoning model

2026-04-02

Global

qwen3.6-plus, qwen3.6-plus-2026-04-02

Qwen3.6-Plus features significant upgrades to its code development capabilities, such as Agentic Coding and frontend programming, and provides a significantly enhanced Vibe Coding experience. Its inference capabilities in generalization scenarios are further enhanced. For multimodal capabilities, its performance in universal object recognition, OCR, and object localization is significantly improved. This version also fixes known issues that were identified after Qwen3.5-Plus was published. The usage method is the same as that of qwen3.5-plus. Overview

Reasoning model

2026-03-04

Global

qwen3.5-flash, qwen3.5-flash-2026-02-23, qwen3.5-122b-a10b, qwen3.5-27b, qwen3.5-35b-a3b

The latest Qwen3.5-Flash model and open-source model from Alibaba, supporting text, image, and video input with fast response speed. Its overall performance is close to qwen3.5-plus. Supports built-in Text generation overview. Overview

Reasoning model

2026-03-04

Global

qwen3.5-plus, qwen3.5-plus-2026-02-15, qwen3.5-397b-a17b

The latest Qwen3.5-Plus model and open-source model from Alibaba, supporting text, image, and video input. It delivers outstanding performance across language understanding, logical reasoning, code generation, agent tasks, image understanding, video understanding, GUI, and other tasks. Supports built-in Text generation overview. Overview

Text-to-image

2026-01-12

Global

wan2.6-t2i

Adds a synchronous API. Supports custom dimensions within total pixel area and aspect ratio constraints. Wan - Text-to-Image V2

Image generation and editing

2026-01-12

Global

wan2.6-image

Supports image editing and mixed image-text output. Wan 2.6: Image generation and editing

Image-to-video - based on first frame

2026-01-12

Global

wan2.6-i2v

Adds multi-shot narrative capability. Supports audio, including automatic dubbing or using custom audio files. Wan image-to-video - based on first frame

Reference-to-video

2026-01-12

Global

wan2.6-r2v

Generates multi-shot videos based on the character and voice from a reference video. Supports automatic dubbing. Wan reference-to-video

Text-to-video

2026-01-12

Global

wan2.6-t2v

Adds multi-shot narrative capability. Supports audio, including automatic dubbing or using custom audio files. Wan text-to-video

Visual understanding

2026-01-12

Global

qwen3-vl-flash, qwen3-vl-flash-2025-10-15

A small-sized visual understanding model from the Qwen3 series that effectively combines thinking and non-thinking modes. It delivers better performance and faster response times than the open source Qwen3-VL-30B-A3B model. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-30b-a3b-thinking, qwen3-vl-30b-a3b-instruct

Based on the new-generation open source Qwen3-VL model, this model is available in thinking and non-thinking versions. It features fast response times and enhanced capabilities for multimodal understanding, inference, visual agent tasks, and extra-long context support for long videos and documents. Its fully upgraded spatial intelligence and object recognition make it suitable for complex real-world tasks. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-8b-thinking, qwen3-vl-8b-instruct

This is an 8B dense open-source model from the Qwen3-VL series, available in thinking and non-thinking versions. It uses less GPU memory and performs multimodal understanding and inference. It supports extra-long context for long videos and documents, 2D/3D visual positioning, and comprehensive spatial intelligence and object recognition. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-32b-thinking, qwen3-vl-32b-instruct

A 32B dense model from the Qwen3-VL series. Its overall performance is second only to the Qwen3-VL-235B model. It excels in document recognition and understanding, spatial intelligence, object recognition, 2D visual detection, and spatial reasoning. It is suitable for complex perception tasks in general scenarios. Image and video understanding

Visual understanding

2026-01-12

Global

qwen3-vl-plus, qwen3-vl-plus-2025-09-23, qwen3-vl-235b-a22b-thinking, qwen3-vl-235b-a22b-instruct

The Qwen3 series visual understanding model effectively combines thinking and non-thinking modes. Its visual agent capabilities are world-class. This version features comprehensive upgrades in visual encoding, spatial intelligence, and multimodal thinking. Its visual perception and recognition capabilities are significantly improved. Image and video understanding

Reasoning model

2026-01-12

Global

qwen3-next-80b-a3b-thinking, qwen3-next-80b-a3b-instruct

A next-generation open-source model based on Qwen3. The thinking modelhas improved instruction following and more concise summaries compared with qwen3-235b-a22b-thinking-2507. See Deep thinking.The instruct modelhas enhanced Chinese comprehension, logical reasoning, and text generation compared with qwen3-235b-a22b-instruct-2507. See Overview.

Reasoning model

2026-01-12

Global

qwen3-max, qwen3-max-2025-09-23

Compared to the qwen3-max-preview version, this release features specialized upgrades in agent programming and tool calling. The official release achieves state-of-the-art (SOTA) performance in its domain and meets the needs of more complex agent scenarios.

Reasoning model

2026-01-12

Global

qwen3-max-preview

The Qwen-Max preview model, built on Qwen3, delivers a substantial improvement in general capabilities over the Qwen2.5 series. It shows significant gains in Chinese and English text understanding, complex instruction following, subjective open-ended tasks, multilingual tasks, and tool calling. It also exhibits fewer knowledge hallucinations. Qwen-Max

Code model

2026-01-12

Global

qwen3-coder-flash, qwen3-coder-flash-2025-07-28

The fastest and most cost-effective model in the Qwen Coder series. Code capabilities (Qwen-Coder).

Code model

2026-01-12

Global

qwen3-coder-plus-2025-09-23

Compared to the previous version (July 22 snapshot), this model has improved robustness for downstream tasks and tool calling, along with enhanced code security. Code generation (Qwen-Coder)

Code model

2026-01-12

Global

qwen3-coder-plus, qwen3-coder-plus-2025-07-22, qwen3-coder-30b-a3b-instruct, qwen3-coder-480b-a35b-instruct

A code generation model based on Qwen3. It has strong Coding Agent capabilities, excels at tool calling and environment interaction, and combines excellent coding skills with general-purpose abilities. Code generation (Qwen-Coder)

Reasoning model

2026-01-12

Global

qwen3-30b-a3b-thinking-2507, qwen3-30b-a3b-instruct-2507

An upgrade of qwen3-30b-a3b. The thinking modelhas improved logical, general, knowledge, and creative capabilities. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Reasoning model

2026-01-12

Global

qwen3-235b-a22b-thinking-2507, qwen3-235b-a22b-instruct-2507

An upgrade of qwen3-235b-a22b. The thinking modelhas significantly improved logical, general, knowledge, and creative capabilities, suitable for challenging reasoning scenarios. See Deep thinking.The instruct modelhas improved creative capabilities and model safety. See Overview.

Reasoning model

2026-01-12

Global

qwen3-235b-a22b, qwen3-30b-a3b, qwen3-32b, qwen3-14b, qwen3-8b

Qwen3 models support both thinking and non-thinking modes. You can switch between the two modes by using the enable_thinking parameter. In addition, the capabilities of Qwen3 models are significantly improved:

1. 1. Reasoning ability: In evaluations for math, code, and logical reasoning, it significantly outperforms QwQ and other non-reasoning models of the same size, reaching top-tier industry performance for its scale.

2. 2. Human preference alignment: Creative writing, role-playing, multi-turn conversation, and instruction-following capabilities are greatly improved. Its general capabilities significantly exceed those of other models of the same size.

3. 3. Agent capabilities: Achieves industry-leading performance in both reasoning and non-reasoning modes and can accurately invoke external tools.

4. 4. Multilingual capabilities: Supports over 100 languages and dialects, with notable improvements in multilingual translation, instruction understanding, and common-sense reasoning.

5. 5. Response format fixes: Fixes response format issues from previous versions, such as abnormal Markdown, mid-response truncation, and incorrect boxed output.

For thinking mode, see Deep thinking. For non-thinking mode, see Overview.

Text extraction

2026-01-12

Global

qwen-vl-ocr-2025-11-20

The Qwen text extraction model. This snapshot version is based on the Qwen3-VL architecture and significantly improves document parsing and text localization capabilities. Text extraction

Text extraction

2026-01-12

Global

qwen-vl-ocr

qwen-vl-ocr is a model specialized for OCR, with significantly improved text extraction from tables, exam papers, and similar images. For details, see Text extraction (Qwen-OCR).

Reasoning model

2026-01-12

Global

qwen-plus-2025-12-01, qwen-plus-2025-09-11

This model is part of the Qwen3 series. Compared to qwen-plus-2025-07-28, it provides enhanced instruction-following capabilities and more concise summaries and responses in thinking mode. See Deep thinking. In non-thinking mode, its Chinese language understanding and logical reasoning capabilities are enhanced. See Overview.

Reasoning model

2026-01-12

Global

qwen-plus-2025-07-28

This model is part of the Qwen3 series. Compared to the previous version, its context length is increased to 1,000,000. For thinking mode, see Deep thinking. For non-thinking mode, see Overview.

Reasoning model

2026-01-12

Global

qwen-plus

Offers a balance of capabilities. Its inference performance, cost, and speed are between those of Qwen-Max and Qwen-Flash, making it suitable for moderately complex tasks.

Multilingual translation

2026-01-12

Global

qwen-mt-lite

A basic-level model from Qwen for text translation. It supports translation between 31 languages. Compared to qwen-mt-flash, it has faster response times and lower costs, making it suitable for latency-sensitive scenarios. Machine translation (Qwen-MT)

Multilingual translation

2026-01-12

Global

qwen-mt-plus, qwen-mt-flash

The Qwen-MT model is a machine translation large language model that is optimized based on the Qwen model. It excels at translation between Chinese and English, between Chinese and other languages, and between English and other languages. The 26 other languages include Japanese, Korean, French, Spanish, German, Portuguese (Brazil), Thai, Indonesian, Vietnamese, and Arabic. In addition to multilingual translation, it provides features such as terminology intervention, domain prompts, and translation memory to improve translation quality in complex application scenarios. See Machine translation (Qwen-MT).

Text generation

2026-01-12

Global

qwen-flash, qwen-flash-2025-07-28

The fastest and most cost-effective model in the Qwen series, suitable for simple jobs.

Type

Time

Deployment scope

Model

Description

Reasoning model

2026-06-18

Japan

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-06-18

Japan

deepseek-v4-pro, deepseek-v4-flash

DeepSeek-V4 series models hosted by Alibaba Cloud (Beijing region). deepseek-v4-pro is a large-scale MoE model with strong general reasoning capabilities. deepseek-v4-flash is a lightweight and cost-effective model optimized for speed.DeepSeek

Reasoning model

2026-06-18

Global

qwen3.7-plus, qwen3.7-plus-2026-05-26

The Qwen3.7-Plus series comprehensively upgrades vision-language capabilities on top of strong text abilities, while maintaining full agent capabilities in coding, tool use, and productivity workflows. Its core feature is multimodal interactive hybrid agent capability, enabling perception of real-world scenes, screen reading and GUI operation, code generation from visual references, and end-to-end mobile app navigation.

Reasoning model

2026-06-18

Global

qwen3.7-max, qwen3.7-max-2026-05-20

Next-generation flagship model in the Qwen Max series. Text-only input, thinking mode enabled by default, supports explicit cache and Function Calling. This model delivers exceptional performance across programming, office productivity, and long-horizon autonomous task execution.

Reasoning model

2026-06-18

Global

glm-5.1

Zhipu GLM-5.1 model, designed for long-range tasks with 200K context support and up to 128K token output. With powerful logical reasoning, long-text understanding, and code generation capabilities, it achieves excellent results on multiple benchmarks, suitable for intelligent interaction, enterprise applications, and development assistance scenarios.GLM - Alibaba Cloud

Text generation and visual understanding

2026-06-18

Global

kimi-k2.5

A visual understanding model from Moonshot AI that excels in general-purpose intelligent tasks such as code generation and visual understanding. It also supports image, video, and text input, along with dialogue and Agent tasks. Kimi - Alibaba Cloud

Reasoning model

2026-06-18

Global

qwen3.6-flash, qwen3.6-flash-2026-04-16

A native vision-language Flash model in the Qwen3.6 series, delivering significant overall performance improvements over Qwen3.5-Flash. Agentic coding capabilities are substantially enhanced (far surpassing the predecessor on multiple code agent benchmarks), along with stronger math and code reasoning. On the vision side, spatial intelligence is notably improved, with object localization and detection being particularly outstanding. Overview

Reasoning model

2026-06-18

Global

qwen3.6-plus, qwen3.6-plus-2026-04-02

Qwen3.6-Plus features significant upgrades to its code development capabilities, such as Agentic Coding and frontend programming, and provides a significantly enhanced Vibe Coding experience. Its inference capabilities in generalization scenarios are further enhanced. For multimodal capabilities, its performance in universal object recognition, OCR, and object localization is significantly improved. This version also fixes known issues that were identified after Qwen3.5-Plus was published. The usage method is the same as that of qwen3.5-plus. Overview