Alibaba claims its new Qwen AI model outperformed ChatGPT and Gemini in coding tests

Alibaba has launched its latest artificial intelligence model, Qwen3.7-Max, claiming the system outperformed several competing AI models from OpenAI and Google in coding benchmarks. According to Alibaba, the model is designed to function as an autonomous coding agent capable of handling long and complex software tasks with limited human input.

The launch highlights China’s growing push to compete with US companies in advanced AI development, especially in the rapidly expanding field of AI coding systems. Chinese technology firms are increasingly investing in models that can operate more independently and perform tasks traditionally handled by software engineers.

What is Qwen3.7-Max?

Unlike standard AI chatbots that mainly answer questions or generate text, Qwen3.7-Max has been designed specifically for “agent-based” coding tasks. This means the model can independently manage workflows, interact with software tools and continue working across long sessions without constant user instructions.

Alibaba says the AI can:

Build front-end software prototypes

Manage large multi-file coding projects

Automate office workflows

Use external tools independently

The company claims the model can operate autonomously for up to 35 hours continuously and reportedly perform more than 1,000 software tool interactions during a single session.

How Alibaba says the AI performed

According to Alibaba, Qwen3.7-Max currently ranks fourth globally on the Code Arena benchmark leaderboard with a score of 1,541. The company says the ranking places the model ahead of some versions of ChatGPT and Gemini in coding-related tasks. Only models from Anthropic reportedly scored higher in the rankings. Alibaba researchers also described an internal experiment where the AI was asked to optimise software for one of the company’s own AI chips. The company claims the model worked continuously for around 35 hours, performing:

432 kernel tests
More than 1,100 software tool calls
Repeated compiling and rewriting of code

According to Alibaba, the AI achieved a tenfold performance improvement over the original implementation despite not being trained on that chip architecture previously.

China’s growing competition with US AI firms

The launch comes at a time when companies such as OpenAI, Google and Anthropic continue to dominate advanced AI development globally. However, Chinese firms are increasingly trying to compete, particularly in the field of autonomous coding agents. Analysts say AI coding systems are becoming one of the most important areas in artificial intelligence because they can potentially automate software development tasks and improve productivity for businesses. Alibaba stated that Qwen3.7-Max supports interfaces compatible with OpenAI and Anthropic systems and can work alongside tools such as Claude Code, OpenClaw and Qwen Code.

Shift from open-source to proprietary models

The launch also marks a strategic shift for Alibaba. Earlier versions of Qwen were released as open-source models, allowing developers to access and modify them freely. However, the new Qwen3.7-Max is proprietary and available only through Alibaba Cloud’s Model Studio API. Industry experts believe this reflects a wider industry trend where companies are moving towards commercial AI platforms as development costs continue to rise.

Benchmark claims still await independent review

Alibaba said Qwen3.7-Max performed strongly across several coding and reasoning benchmarks and came close to Anthropic’s Claude Opus 4.6 Max in multiple tests. However, many of these benchmark results were self-reported by the company.

The launch of Qwen3.7-Max shows how competition in the global AI race is accelerating. Companies are now moving beyond simple chatbots and focusing on systems that can independently perform complex tasks over extended periods. As AI coding tools continue improving, analysts believe the technology could significantly change software development, engineering workflows and enterprise automation in the coming years.