Alibaba has unveiled Qwen4-Max, its new flagship AI model, in the latest challenge from a Chinese lab to the US frontier leaders. Announced on Sunday 15 November, the model is built on a new Qwen4 architecture: a 2.4-trillion-parameter mixture-of-experts backbone with 71 billion parameters activated per token, an additional 328 billion parameters devoted to N-gram embeddings, native vision capabilities, and a one-million-token context window.

Alibaba pitches Qwen4-Max as an AI teammate for professional work rather than a chatbot: 3D modeling, game development, precision manufacturing, finance, law, healthcare and frontier science. On its own Work360 benchmark of real industry tasks, the company reports a score of 52.69, up from 47.74 for the previous Qwen3.8-Max. It also claims big efficiency gains: token usage down by roughly 24–35% on Work360 and about 48–49% on Terminal-Bench 4.0, with the average cost per task falling from $4.43 to $1.67 — a 62% cut — and lower API prices than its predecessor. Those are Alibaba's own numbers; an independent testing site, Vals, currently places Qwen4-Max 11th of 38 models on its overall index, a useful dose of scepticism beside the launch claims.

The launch blog leans heavily on long-horizon work: building playable games in Unity, Unreal Engine and Godot from a single brief, planning automotive production lines and machining parts in real factory settings — Alibaba says the manual effort to program a single workpiece fell from about 475 minutes to 24 minutes — and compiling an 86-page pre-IPO legal due-diligence report from 198 inconsistently formatted working papers. On safety, Alibaba says training included security checks alongside coding tests to cut risks such as code injection and broken access controls, safeguards against misuse in cybercrime and hazardous materials were strengthened, and internal testing found no evidence of significant high-risk capabilities.

Qwen4-Max is not open-weight, unlike much of the Qwen family — Alibaba says smaller Qwen4 models will be released as open weights soon. The model is available through Qwen's API and chat platform. The launch caps a busy month for Chinese labs, following Tencent's Hunyuan 4 and Moonshot's Kimi K3.1 Pro, and Alibaba claims Qwen4-Max scores near the top of coding benchmarks, ahead of Anthropic's Claude Fable 5.1 on the tests it highlights.

Sources