AI BriefingQwen3.6-Max-Preview: Next-Generation Model with Enhanced Agentic Coding and World Knowledge
Key point
The successor preview model to Qwen3.6-Plus, with improved coding, knowledge, and tool-calling.
Details
Released as the successor to Qwen3.6-Plus, this preview model shows overall improvements in agentic coding, world knowledge, and instruction following performance.
It showed strength across major coding benchmarks, with confirmed improvements of SkillsBench +9.9, SciCode +6.3, NL2Repo +5.0, and Terminal-Bench 2.0 +3.8. It reportedly achieved the top score on 6 major coding benchmarks.
Knowledge performance also improved, with SuperGPQA +2.3 and QwenChineseBench +5.3, and instruction-following ability improved by ToolcallFormatIFBench +2.8.
For agentic tasks, it supports the preserve_thinking feature, which allows reasoning content from previous turns to be preserved in messages, and with enable_thinking: True, reasoning content and responses can be received separately via streaming.
It is provided as a hosted proprietary model based on Alibaba Cloud Model Studio, and also supports an OpenAI-compatible chat completions / responses API and an Anthropic-compatible interface. The API model name is qwen3.6-max-preview, and regional Base URLs are provided for Beijing, Singapore, and the US (Virginia).
Currently, interactive testing is available in Qwen Studio, and it was announced that it will soon be available via the Model Studio API.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Toss Builds 'Toss Benchmark' to Bridge Gap Between Public Leaderboards and Real-World Performance
PPM Framework Addresses Sparse Reward Inefficiency in Long-Horizon LLM Reasoning