Model Studio models-Hologres(Hologres)-阿里云帮助中心

This guide shows how to integrate Alibaba Cloud Model Studio with Hologres. You can deploy models from Model Studio in Hologres by using an API key and call them through AI Functions. This approach lets you perform AI development without moving data out of the database.

Introduction

Alibaba Cloud Model Studio (Bailian) is a one-stop platform for large language model development and application. It integrates Qwen and mainstream third-party models, providing developers with OpenAI-compatible APIs, end-to-end model services, and visual application building capabilities. Model Studio provides ready-to-use model services that let you call the full Qwen model lineup without deploying or maintaining infrastructure.

Hologres is deeply integrated with Alibaba Cloud Model Studio (Bailian). You can deploy Model Studio models in Hologres with an API key and invoke them using AI Function. This lets you develop and build AI applications without moving data out of your database.

Billing

Network fees: The supported regions for Model Studio are China (Beijing) and Singapore. Accessing Model Studio from a Hologres instance may incur network fees. Network fees are waived during the beta period, and the billing start date will be announced on the official website.
Model invocation fees: Model Studio charges these fees based on API call volume. For details, see model pricing and the Model Studio console.

Limitations

Supported instance versions: Hologres V4.0.18 and later, or V4.1.2 and later.
Supported regions:
- China (Beijing, Ulanqab)
- China (Hangzhou, Shanghai)
- China (Shenzhen)
- Singapore

Model list and parameters

Deploy a model

In the Hologres console, go to Instances. Select your target instance and click AI Model at the top of the instance details page. On the Models page, you can deploy a Model Studio model. Select Alibaba Cloud Model Studio as the provider and configure the following parameters. The key parameters include:

Model Type: The Model Studio model you want to deploy. For details, see the Model list section below. Models that are not on this list are not supported.
API Key: You must activate Alibaba Cloud Model Studio and obtain an API key before using the service. Enter this key during deployment. For instructions on how to obtain an API key, see Get an API Key.
Model Parameter Settings: After selecting a model type, configure its parameters to meet your business needs. For details, see the Parameter description section below. You can also configure the retry mechanism.

Parameters

The following parameters are supported by different model types. For a complete description, see the Model Studio console and API documentation.

Text models:
- max_tokens: The maximum number of tokens to return. For the model's token limit, see the official Model Studio documentation.
- temperature: The sampling temperature, which controls the diversity of the generated output. Valid values: [0, 2.0).
- top_p: The probability threshold for nucleus sampling. Valid values: (0, 1.0]. Both temperature and top_p control diversity. We recommend setting only one of them.
- Qwen-Omni series: In addition to the common text parameters, this series supports modalities (specifies the output as text or audio), audio.voice (specifies the voice for audio output), and audio.format (specifies the audio format; wav is a supported format).

Translation models: Configure the following parameters to improve translation quality. For complete usage details, see Translation models.

source_lang: The source language. For details, see the language list.
terms: A JSON-formatted list of custom translation terms.
tm_list: The translation memory, which provides example source-target sentence pairs in JSON format.
domains: Text-based, domain-specific prompts.

Example:

{
  "extra_body": {
    "translation_options": {
      "source_lang": "zh", 
      "domains": "The sentence is from Ali Cloud IT domain. ", 
      "terms": [
        {"source": "生物传感器", "target": "biological sensor"},
        {"source": "身体健康状况", "target": "health status of the body"}
      ], 
      "tm_list":[
        {"source": "您可以通过如下方式查看集群的内核版本信息:", "target": "You can use one of the following methods to query the engine version of a cluster:"},
        {"source": "bla", "target": "bla"}
      ]
    }
  }
}

Embedding models: dimension specifies the vector dimension. You can modify this parameter for only some models. For detailed usage instructions, see Vectorization models.
- text-embedding-v4 supports 2,048, 1,536, 1,024 (default), 768, 512, 256, 128, and 64.
- text-embedding-v3 supports 1,024 (default), 768, 512, 256, 128, or 64.
- qwen3-vl-embedding supports 2,560 (default), 2,048, 1,536, 1,024, 768, 512, and 256.

Retry mechanism

When you deploy a model, you can configure the retry mechanism for failed calls. The following parameters are available:

max_retries: The maximum number of retries. Default value: 2. Value range: [0, 100].
initial_retry_delay: The initial delay before a retry, in seconds. Default value: 0.5. Value range: [0.5, 8].
max_retry_delay: The maximum delay for a retry, in seconds. Default value: 8. Value range: [1, 60].
timeout: The timeout for each request, in seconds. Default value: 600. Value range: [1, 1200].

Model list

Model Studio offers models for text generation, translation, embedding, and multimodal tasks. The following table lists the model category, model_type, task type, input/output, notes, and cross-region support for each model.

Model category	`model_type`	Task type	Input/output	Notes	Cross-region support
Text generation	qwen3.6-plus	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-max	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-max-2026-01-23	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-max-preview	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-max	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-max-latest	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-plus	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-plus-latest	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-flash	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-long	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-long-latest	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwq-plus	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwq-plus-latest	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-v3.2	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-v3.2-exp	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-v3.1	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1-0528	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-v3	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1-distill-qwen-1.5b	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1-distill-qwen-7b	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1-distill-qwen-14b	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	deepseek-r1-distill-qwen-32b	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	kimi-k2-thinking	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	Moonshot-Kimi-K2-Instruct	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	glm-4.6	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	glm-4.7	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	glm-5	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	MiniMax-M2.1	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	MiniMax-M2.5	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	MiniMax/MiniMax-M2.1	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	MiniMax/MiniMax-M2.5	chat/completions	Text input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-235b-a22b-instruct	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-235b-a22b-thinking	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-32b-instruct	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-32b-thinking	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-8b-instruct	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-8b-thinking	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-plus	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-vl-flash	chat/completions	Image/video input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-vl-ocr	chat/completions	Image input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen-vl-ocr-latest	chat/completions	Image input, text output	Supports parameters: `temperature`, `top_p`, and `max_tokens`.	Yes
	qwen3-omni-flash	chat/completions	Text/image/audio/video input, text/audio output	Supports parameters: `temperature`, `top_p`, `max_tokens`, `modalities`, and `audio`.	Yes
Translation	qwen-mt-plus	translation	ai_translate	Supports parameters: `source_lang`, `terms`, `tm_list`, and `domains`.	Yes
	qwen-mt-flash	translation	ai_translate	—	Yes
	qwen-mt-turbo	translation	ai_translate	—	Yes
	qwen-mt-lite	translation	ai_translate	—	Yes
Embedding	text-embedding-v1	embedding	`ai_embed`, text input, `float[]` output	Embedding dimensions: 1,536	Yes
	text-embedding-v2	embedding	`ai_embed`, text input, `float[]` output	Embedding dimensions: 1,536	Yes
	text-embedding-v3	embedding	`ai_embed`, text input, `float[]` output	Embedding dimensions: 2,048, 1,536, 1,024 (default), 768, 512, 256, 128, or 64	Yes
	text-embedding-v4	embedding	`ai_embed`, text input, `float[]` output	Embedding dimensions: 1,024 (default), 768, 512, 256, 128, or 64	Yes
	tongyi-embedding-vision-plus	embedding	`ai_embed`, text/image/video input, `float[]` output	Embedding dimensions: 1,152. Video input is only supported in the China (Beijing) and Singapore regions.	Yes
	tongyi-embedding-vision-flash	embedding	`ai_embed`, text/image/video input, `float[]` output	Embedding dimensions: 768. Video input is only supported in the China (Beijing) and Singapore regions.	Yes
	multimodal-embedding-v1	embedding	`ai_embed`, text/image/video input, `float[]` output	Embedding dimensions: 1,024. Video input is only supported in the China (Beijing) and Singapore regions.	Yes
	qwen3-vl-embedding	embedding	`ai_embed`, text/image/video input, `float[]` output	Embedding dimensions: 2,560 (default), 2,048, 1,536, 1,024, 768, 512, or 256	Yes
Image generation and editing	qwen-image-2.0-pro	image-generation	Supports image editing and text-to-image generation.	—	Yes
	qwen-image-2.0-pro-2026-03-03	image-generation	Supports image editing and text-to-image generation.	—	Yes
	qwen-image-2.0	image-generation	Supports image editing and text-to-image generation.	—	Yes
	qwen-image-2.0-2026-03-03	image-generation	Supports image editing and text-to-image generation.	—	Yes
	qwen-image-max	image-generation	Supports text-to-image generation.	—	Yes
	qwen-image-plus	image-generation	Supports text-to-image generation.	—	Yes
	qwen-image	image-generation	Supports text-to-image generation.	—	Yes
	qwen-image-edit	image-generation	Supports image editing.	—	Yes
	qwen-image-edit-plus	image-generation	Supports image editing.	—	Yes
	qwen-image-edit-max	image-generation	Supports image editing.	—	Yes
	wan2.7-image-pro	image-generation	Text-to-image (single images only); Supports 4K HD output.	—	Yes
	wan2.7-image	image-generation	Supports text-to-image generation.	—	Yes
Video generation	wan2.6-r2v	video-generation	Reference-to-video (r2v) generation	—	Yes
	wan2.6-r2v-flash	video-generation	Reference-to-video (r2v) generation	—	Yes
	wan2.6-t2v	video-generation	Text-to-video (t2v) generation	—	Yes
	wan2.6-i2v-flash	video-generation	Image-to-video (i2v) generation from the first frame	—	Yes
	wan2.6-i2v	video-generation	Image-to-video (i2v) generation from the first frame	—	Yes
	wan2.2-kf2v-flash	video-generation	Keyframe-to-video (kf2v) generation from the first and last frames	—	Yes
	wan2.7-t2v	video-generation	Text-to-video generation	—	Yes
	wan2.7-i2v	video-generation	Multimodal input (text, image, video, or audio) to video output	—	Yes
	wan2.7-r2v	video-generation	Multimodal input (text, image, video, or audio) to video output	—	Yes
	wan2.7-videoedit	video-generation	Video generation and editing from multimodal input (text, image, or video).	—	Yes
	happyhorse-1.0-t2v	video-generation	Text-to-video generation	Only supported in China (Hangzhou) and China (Shenzhen).	Yes
	happyhorse-1.0-i2v	video-generation	Image-to-video (i2v) generation from the first frame	Only supported in China (Hangzhou) and China (Shenzhen).	Yes
	happyhorse-1.0-r2v	video-generation	Reference-to-video (r2v) generation	Only supported in China (Hangzhou) and China (Shenzhen).	Yes
	happyhorse-1.0-video-edit	video-generation	Video editing	Only supported in China (Hangzhou) and China (Shenzhen).	Yes
Speech-to-text	fun-asr	speech-to-text	Speech recognition	—	Yes

Model usage

Once a model is deployed, you can use an AI Function in Hologres to call it. This lets you perform inference and build AI applications without moving data from your database. For usage instructions, see the AI Function List. For a best practice example, see Best Practice: Use AI Functions to Build a High-Performance Image Analysis System for Autonomous Driving.