Alibaba releases Qwen3.8-Max with 2.4 trillion parameters and autonomous coding agent

Alibaba isn't easing into the AI arms race this week — it's cannonballing in. The company just unveiled Qwen3.8-Max, a 2.4 trillion parameter model that dwarfs most rivals in raw size, paired with a coding agent Alibaba claims can work completely unsupervised for more than ten days at a stretch, starting from a blank folder and ending with a finished product.

That unsupervised claim is the real headline here, not the parameter count. Plenty of labs have chased bigger and bigger models for years, and the returns on pure scale have gotten murkier. What's different is the idea of an agent that doesn't need a human checking in every few hours to keep it on track. Ten-plus days of autonomous coding, if it holds up under real-world testing, would put Qwen3.8-Max well ahead of the typical agent workflows people are using today, most of which still need frequent human course-correction.

And Alibaba isn't keeping this locked behind an API wall. The plan is to release Qwen3.8-Max as open-source next week, letting developers, researchers, and probably a few competitors pull the model apart to see how it actually performs outside a controlled demo. That's a meaningfully different strategy than the closed, subscription-gated approach OpenAI and Anthropic have leaned on, and it fits the pattern Alibaba's Qwen team has followed for a while now — build big, then hand it out.

There's an obvious catch worth watching: bold uptime claims for autonomous agents have a habit of not surviving contact with messy, real production codebases. Ten days without supervision sounds great until the agent wanders off into a design decision nobody wanted. Alibaba's willingness to open-source the thing means outside developers will get to stress-test that claim fast, for better or worse.