Text Generation and Visual Understanding

Qwen Plus, Max, and Flash accept text, images, and video and respond with text. Use the same model to generate content, understand images, or analyze video. Image or video input alone does not require switching to a VL model.

Choose a model

Plus, Max, and Flash accept text, images, and video and respond with text. Text generation creates or processes text based on your input. Image and video understanding interprets visual content and answers questions, including tasks that combine visual input with text instructions.

• Prioritize quality: qwen4-max-preview (Token Plan only) for tasks with higher reasoning and understanding requirements.
• Balanced: qwen3.7-plus balances capability and cost.
• Prioritize cost: qwen3.8-flash; verify that results meet your requirements.

For third-party models, consider kimi-k3 for long-running coding tasks and glm-5.3 for coding and agent tasks. Check their input modalities, tools, and regional availability separately; the Qwen capabilities below do not apply automatically.

Get started with Text generation and Image and video understanding.

Deep thinking: Use step-by-step reasoning for problems that require multiple stages of analysis, allowing the model to reason more thoroughly before answering. Adjust the thinking mode to suit your task. See Deep thinking.

Tool calling: Models can call tools based on your input. Function Calling connects functions or business APIs that you define. Built-in tools provide web search, a code interpreter, and web scraping to retrieve external information or process data. See Tool calling.

Structured output: Organize information from text, images, or video into valid JSON for programs to read, validate, and process. See Structured output.

Batch inference: For large numbers of requests that do not require immediate results, batch inference can reduce request costs. See Batch inference.

Domain-specific models

For specific tasks, you can also choose a model optimized for that task.

• Translation: Qwen-MT translates text across languages.
• Long documents: Qwen-Long processes very long documents through file uploads and references.
• Role-playing: Qwen-Character supports human-like conversations that maintain a defined persona.
• Data mining: Qwen-Doc extracts information, classifies content, and generates summaries from documents.
• Deep research: Qwen-Deep-Research plans research steps and uses multiple rounds of web search to produce research reports.
• Dialogue analysis: Tongyi-Xiaomi extracts dialogue information, classifies conversations, and analyzes satisfaction.
• GUI interaction: Perform interface operations based on screenshots and instructions.
• Text extraction: Qwen-OCR extracts text from images of documents, tables, or handwriting.

Regions and invocation methods vary by model. See the linked guides for details.

Recommended models

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen4-max-preview	1M	Supported	Supported	Supported	Supported
qwen3.8-max	1M	Supported	Supported	Supported	Supported
qwen3.8-flash	1M	Supported	Supported	Supported	Supported
qwen3.7-plus	1M	Supported	Supported	Supported	Supported
deepseek-v4.1-flash-1017	1M	Supported	Supported	Supported	Supported
glm-5.3	1M	Supported	Supported	Unsupported	Supported
kimi-k3.1-pro	1M	Supported	Supported	Supported (Responses)	Supported (JSON Mode)
kimi-k3	1M	Supported	Supported	Unsupported	Supported
MiniMax-M3.1-Flash	1M	Supported	Supported	Unsupported	Unsupported
mimo-v2.6-pro-ultraspeed	1M	Supported	Supported	Unsupported	Supported
step-5-preview	1M	Supported	Supported	Unsupported	Supported

Image and video parameters

Model ID	Context	Max output	Max pixels/image	Max video duration	Max video size	Max images	Max videos	Function calling	Built-in tools	Structured output
qwen4-max-preview	1M	393,216 (thinking + output)	16M	2 hours	2 GB	2048	64	Supported	Supported	Supported
qwen3.8-max	1M	128k	16M	2 hours	2 GB	2048	64	Supported	Supported	Supported
qwen3.8-flash	1M	128k	16M	2 hours	2 GB	2048	64	Supported	Supported	Supported
qwen3.7-plus	1M	128k	16M	2 hours	2 GB	2048	64	Supported	Supported	Supported

Older models

These models are no longer the first recommendation. Each model is listed once, without duplicate preview names or snapshots.

View older models

Qwen3.7

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen3.7-max	1M	Supported	Supported	Supported	Supported
qwen3.7-flash	1M	Supported	Supported	Supported	Supported

Image and video parameters

Model ID	Context	Max pixels/image	Max video duration	Max video size	Max images	Max videos	Function calling	Built-in tools	Structured output
qwen3.7-flash	1M	16M	2 hours	2 GB	256	64	Supported	Supported	Supported

Qwen3.6

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen3.6-plus	1M	Supported	Supported	Supported	Supported
qwen3.6-flash	1M	Supported	Supported	Supported	Supported

Qwen3.5

Image and video parameters

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen3.5-plus	1M	Supported	Supported	Supported	Supported
qwen3.5-flash	1M	Supported	Supported	Supported	Supported
qwen3.5-397b-a17b	256k	Supported	Supported	Supported	Supported
qwen3.5-122b-a10b	256k	Supported	Supported	Supported	Supported
qwen3.5-27b	256k	Supported	Supported	Supported	Supported
qwen3.5-35b-a3b	256k	Supported	Supported	Supported	Supported

Qwen3-Coder

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen3-coder-flash	1M	Supported	Supported	Unsupported	Supported

Qwen-MT

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen-mt-plus	16k	Unsupported	Unsupported	Unsupported	Unsupported
qwen-mt-flash	16k	Unsupported	Unsupported	Unsupported	Unsupported
qwen-mt-lite	16k	Unsupported	Unsupported	Unsupported	Unsupported

Qwen-Long

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen-long	10M	Unsupported	Unsupported	Unsupported	Supported

Qwen-Character

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen-plus-character	32k	Unsupported	Unsupported	Unsupported	Unsupported
qwen-plus-character-ja	32k	Unsupported	Unsupported	Unsupported	Unsupported
qwen-flash-character	8k	Unsupported	Unsupported	Unsupported	Unsupported

Qwen

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
qwen-plus and its snapshots	1M	Supported	Supported	Supported	Supported
qwen-max and its snapshots	128k	Unsupported	Supported	Supported	Supported
qwen-flash and its snapshots	1M	Supported	Supported	Supported	Supported

DeepSeek

Feature support

Model ID	Context	Thinking mode	Function Calling	Built-in tools	Structured output
deepseek-r1-distill-llama-70b	128k	Supported	Unsupported	Unsupported	Unsupported
deepseek-r1-distill-qwen-1.5b	128k	Supported	Unsupported	Unsupported	Unsupported