Alibaba says its new AI beats ChatGPT, Gemini in coding
28 May 2026

Alibaba has unveiled its latest artificial intelligence (AI) model, Qwen3.7-Max, which has outperformed OpenAI and Google's models in coding benchmarks.

The advanced AI system is specifically designed for autonomous coding tasks and can operate without human intervention for as long as 35 hours.

It has already secured a spot in the top five globally on Code Arena rankings with a score of 1,541.

Qwen3.7-Max is designed for agent-based tasks
Advanced features

Unlike conventional chatbots that mainly provide answers to queries, Qwen3.7-Max is designed for "agent-based" tasks.

This means it can independently manage complex workflows instead of just answering one prompt at a time.

The model can build front-end prototypes, manage large multi-file software projects, automate office tasks with external tools, and run autonomously for long periods without human input.

How the AI performed during tests
Coding prowess

Alibaba researchers recently tested Qwen3.7-Max by optimizing code for one of the company's own AI chips.

The model worked continuously for some 35 hours, running 432 kernel tests and making over 1,100 tool calls while compiling, measuring, and rewriting code on its own.

The AI achieved a 10x performance improvement over the original implementation despite reportedly never having seen that chip architecture during training.

Qwen3.7-Max is not open-source
Strategic move

The launch of Qwen3.7-Max also marks a strategic shift for Alibaba as previous Qwen models were open-sourced, but the latest Max version is proprietary and available only through Alibaba Cloud's Model Studio API.

Beyond coding, the model can also monitor AI training systems, detect suspicious behavior during software engineering tests, and guide robots through physical spaces using paired navigation systems.