New models from DeepSeek and Alibaba are competing with Anthropic and OpenAI, according to reports.
Alibaba is making the weights of its newest AI model, Qwen3.8-Max, freely available, the company stated.
The Qwen team announced that Qwen3.8-Max delivers 'comprehensive improvements across coding, work, research, and long-horizon tasks'.
The Qwen3.8-Max model can operate unsupervised for days while delivering results.
Completely on its own for about five days or 125 hours of continuous effort,
Alibaba reports that in one test, the model rebuilt and improved a maths research paper's experiment from scratch during 125 hours of continuous effort.
Number of human teams beaten by the Qwen model in a 24-hour contest against 526 human teams.
In a live online contest with a 24-hour deadline involving 526 human teams, the Qwen model beat all but 68 teams.
Competition with Anthropic tools
Anthropic's Claude Opus 5, launched on 24 July, is designed to run unsupervised for long stretches.
Anthropic's Claude Code tool can plan fixes, read software, and test work with minimal human input.
Additionally, Anthropic's Claude Cowork is designed for users to hand off long-duration tasks, such as spreadsheets and research, for the AI to finish largely on its own.
There is a growing AI rivalry between the US and China.
The US government has spent the past two years restricting advanced chip exports to Chinese AI firms to keep companies such as Alibaba a generation behind.
Neither the performance figures from Alibaba nor Anthropic have been verified by independent researchers.
