Alibaba's Qwen3.7-Max Scores Highest Ever for a Chinese AI Model

Alibaba unveiled Qwen3.7-Max on May 20 at the Alibaba Cloud Summit in Hangzhou, marking a significant milestone for Chinese AI development. The model scored 56.6 on the Artificial Analysis Intelligence Index at launch, placing it between Gemini 3.5 Flash and Claude Opus 4.7 globally — the highest-ranked Chinese AI model ever recorded on that leaderboard, currently sitting in the top 10 of 151 measured models.

Qwen3.7-Max is purpose-built for agentic workflows, supporting autonomous execution across hundreds or thousands of steps. It leads on HMMT mathematics benchmarks at 97.1%, completed a remarkable 35-hour autonomous coding session, and features a 1 million token context window. On coding benchmarks, it tops Opus 4.6 on Terminal-Bench, SWE-Bench Pro, and MCP-Atlas. Pricing is aggressively competitive at $2.50 per million input tokens and $7.50 per million output tokens — roughly a third of comparable Western models.

The model is closed-weights and API-only, accessible through Alibaba Cloud's DashScope and Model Studio platforms, plus third-party aggregators like OpenRouter. A companion model, Qwen3.7-Plus, adds vision input capabilities with a Vision Arena ranking of #16. In a notable demonstration, Qwen3.7-Max operated a robot dog through tool-use calls via the Qwen-RobotClaw harness and Qwen-RobotNav navigation foundation model, showcasing physical world agent capabilities that go beyond purely digital tasks.

For developers and enterprises in MENA looking for alternatives to Western frontier models, Qwen3.7-Max offers competitive performance at significantly lower cost, with the added advantage of multilingual Arabic support in the Qwen model family.