Qwen3.8 Max

Multimodal reasoning and agentic coding with 1M-token context.

Playground
History
APIPricing
~0.00s

Chat

0 messages

Press Enter to send, Shift + Enter for new line • Max 5 files (10MB each)

Qwen3.8 Max: Multimodal Reasoning and Agentic Coding LLM

What is Qwen3.8 Max?

Qwen3.8 Max is the flagship large language model from Alibaba's Qwen team and the most capable model in the Qwen family to date. It is a mixture-of-experts (MoE) system with 2.4 trillion total parameters that activates roughly 95 billion parameters per request, built on the Qwen3.5 architecture. The model accepts text and image input and returns text, with a context window of up to 1 million tokens. It is designed for complex reasoning, visual understanding, coding, and long-horizon agentic workflows — the kind of work where a model has to plan, act, verify, and iterate over many steps rather than answer a single prompt.

Key Features

  • •1M-token context for whole-repository code, long documents, and multi-file research.
  • •Multimodal input — combine a text prompt with an image for vision-grounded reasoning.
  • •Mixture-of-experts efficiency — ~95B active parameters keep inference tractable at frontier scale.
  • •Long-horizon agentic execution — planning, tool use, self-verification, and iteration.
  • •Large output budget — up to 131,072 tokens per response.
  • •OpenAI- and Anthropic-compatible request formats for easy integration.

Best Use Cases

Qwen3.8 Max shines on repository-scale coding agents, autonomous refactors, and bug-fixing loops that run across many steps. Its long context makes it strong for long-document analysis, legal and financial review, knowledge bases, and multi-step research assistants. The multimodal input supports document understanding, screenshot reasoning, and visual QA. Teams building agent frameworks can drop it into existing OpenAI/Anthropic-compatible clients.

Prompt Tips and Output Quality

Give the model explicit, structured, multi-part instructions — it follows numbered requirements closely. In hands-on testing, a senior-engineer coding prompt produced a correct, well-commented Kadane's-algorithm implementation with accurate complexity analysis and passing test cases, finishing cleanly without truncation. For agentic runs, break goals into checkpoints so intermediate work can be inspected. For pure text tasks, leave the image field empty; add an image only when the task is vision-grounded.

FAQs

Is Qwen3.8 Max multimodal? Yes — it accepts text and image input and returns text.

What context length does it support? Up to 1 million tokens, suited to whole repositories and long documents.

What is it best at? Agentic coding, long-horizon execution, long-context reasoning, and multimodal understanding.

Who makes Qwen3.8 Max? Alibaba's Qwen team; it is the flagship of the Qwen3.8 series.

How do I call it? Send a prompt to the API endpoint using an OpenAI-style messages request and read the generated text from the response.