Alibaba Cloud Unveils Qwen 3.6-Max-Preview and Other Qwen 3.6 Series Models

Alibaba Cloud has unveiled Qwen 3.6-Max-Preview, its most advanced AI model, signaling a strategic pivot from open-source to a proprietary, hosted system. The model tops multiple benchmarks, including SWE-bench Pro for real-world software engineering, Terminal-Bench 2.0 for command-line execution, and SkillsBench for general problem-solving. It also outperformed competitors like Claude on ToolcallFormatIFBench for instruction-following. Performance gains include a 2.3% improvement on SuperGPQA for reasoning and a 5.3% boost on QwenChineseBench for Chinese language tasks. The model supports a 256k token context window and introduces a preserve_thinking feature, which maintains reasoning traces across interactions—ideal for complex agent-driven workflows.

This release marks a notable departure from Alibaba's earlier open-access strategy. While lower-tier models like Qwen 3.6-35B-A3B remain open source, Max-Preview is a paid, controlled offering available via Qwen Studio and Alibaba Cloud's Model Studio API, supporting both OpenAI and Anthropic API formats for easy integration. The shift comes amid broader industry trends toward monetization, with Alibaba recently shutting down the free tier of Qwen Code and MiniMax tightening its open-source license. Independent analysis from Artificial Analysis ranks Qwen 3.6-Max-Preview as the second-best performer behind Muse Spark, placing it well above average for reasoning models in its pricing category. The model is still under active development and currently limited to text input, with no image capabilities at launch.