Following the release of Qwen3.6-Plus, we are excited to share an early preview of our next-generation flagship model: Qwen3.6-Max-Preview. Compared to Qwen3.6-Plus, this preview version delivers enhanced world knowledge and instruction-following capabilities, along with significantly improved performance in agent programming across multiple benchmarks. As a preview release, the model is still under active iteration, and future versions will continue to be optimized.
The key features of Qwen3.6-Max-Preview include:
- Significantly enhanced agent programming capabilities compared to Qwen3.6-Plus.
- Stronger world knowledge and improved instruction following.
- Superior performance in real-world agent scenarios and knowledge reliability.
You can engage in interactive conversations via Qwen Studio (https://chat.qwen.ai/), and it will soon be available through the Alibaba Cloud Bailian API under the name qwen3.6-max-preview.
Model Performance
The following section presents a benchmark comparison between Qwen3.6-Max-Preview and leading frontier models. Compared to Qwen3.6-Plus, the preview version achieves significant improvements in agent programming (e.g., SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, Terminal-Bench 2.0 +3.8), demonstrates stronger world knowledge (SuperGPQA +2.3, QwenChineseBench +5.3), and exhibits better instruction following (ToolcallFormatIFBench +2.8).
Getting Started
Qwen3.6-Max-Preview
Qwen3.6-Max-Preview will be available via the Alibaba Cloud Bailian API under the model name qwen3.6-max-preview. You can also experience it instantly on Qwen Studio.
API Updates
Alibaba Cloud Bailian supports industry-standard protocols, compatible with OpenAI-spec chat completions and responses APIs, as well as Anthropic-compatible API interfaces.
This release supports the preserve_thinking feature: it retains thinking content from all previous turns in the message, recommended for agent tasks.
Summary
Qwen3.6-Max-Preview is an early preview of our next-generation flagship model, showing significant improvements over Qwen3.6-Plus in agent programming, world knowledge, and instruction following. It achieves top scores on six major programming benchmarks—SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, and SciCode—marking a substantial leap forward from its predecessor. Simultaneously, it demonstrates superior performance in knowledge (SuperGPQA, QwenChineseBench) and instruction following (ToolcallFormatIFBench).
As a preview version, Qwen3.6-Max-Preview is still under active development. We will continue to iterate on the model, bringing further improvements in subsequent versions. We welcome feedback from the community and look forward to seeing your creations. Stay tuned!