Alibaba releases Qwen3.6-Max preview version

Zhitong
2026.04.20 23:45

Alibaba released the Qwen3.6-Max preview version, which is a follow-up to Qwen3.6-Plus, featuring stronger world knowledge and instruction-following capabilities, with significant improvements in agent programming performance. The preview version surpassed mainstream models in multiple benchmark tests, becoming the best domestic model. Qwen3.6-Max-Preview will be provided through Alibaba Cloud's Bai Lian API, and users can experience it on Qwen Studio

According to Zhitong Finance APP, following the release of Qwen3.6-Plus, Alibaba (09988) Qwen large model team shared an early preview of the next-generation flagship model: Qwen3.6-Max-Preview. Compared to Qwen3.6-Plus, this preview version brings stronger world knowledge and instruction-following capabilities, as well as significantly improved agent programming performance across multiple benchmarks. As a preview version, the model is still actively iterating, and subsequent versions will continue to optimize. The main features of Qwen3.6-Max-Preview include: significantly enhanced agent programming capabilities compared to Qwen3.6-Plus; stronger world knowledge and instruction-following; and better performance in real-world scenarios regarding agent and knowledge reliability.

Model Performance

The following shows the evaluation comparison of Qwen3.6-Max-Preview with mainstream cutting-edge models. Compared to Qwen3.6-Plus, the preview version achieved significant improvements in agent programming (e.g., SkillsBench +9.9, SciCode +10.8, NL2Repo +5.0, Terminal-Bench 2.0 +3.8), stronger world knowledge (SuperGPQA +2.3, QwenChineseBench +5.3), and better instruction-following (ToolcallFormatIFBench +2.8). In the recently updated large model ranking by the authoritative third-party evaluation Artificial Analysis, Qwen3.6-Max-Preview's performance surpassed models like GLM5.1 and MiniMax-M2.7, topping the list of the best domestic models.

Getting Started with Qwen3.6-Max-Preview

Qwen3.6-Max-Preview will be provided through Alibaba Cloud's BaiLian API, with the model name qwen3.6-max-preview. You can also experience it instantly on Qwen Studio.

API Updates

Alibaba Cloud's BaiLian supports industry-standard protocols, compatible with OpenAI specifications for chat completions and responses APIs, as well as compatible with Anthropic's API interface.

This release supports the preserve_thinking feature: retaining all previous rounds of thought content in messages, recommended for agent tasks.

Summary

Qwen3.6-Max-Preview is an early preview of the next-generation flagship model of the Qwen large model, with significant improvements in agent programming, world knowledge, and instruction-following compared to Qwen3.6-Plus. It achieved the highest scores in six major programming benchmarks—SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, and SciCode have made significant improvements compared to their predecessors. They also perform better in knowledge (SuperGPQA, QwenChineseBench) and instruction adherence (ToolcallFormatIFBench).

As a preview version, Qwen3.6-Max-Preview is still under active development. The Qwen large model team will continue to iterate on the model, with subsequent versions bringing further enhancements.

Currently, the Qwen3.6 series has launched Qwen3.6-Max-Preview, Qwen3.6-Plus, Qwen3.6-Flash, and has open-sourced Qwen3.6-35B-A3B. Among them, Qwen3.6-Plus topped the daily, weekly, and trending charts on OpenRouter, the world's largest third-party API calling platform, while Qwen3.6-35B-A3B topped the HuggingFace global open-source large model leaderboard, making Qwen3.6 the most popular large model among enterprises and developers at present

Source: Zhitong The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.