Alibaba opened a Max-class Qwen for the first time — 2.4 trillion parameters, and a licence that is not Apache
Alibaba published the weights of Qwen3.8-Max on Hugging Face at 10:00 Beijing time today, the first time a Max-class Qwen model has been opened. It carries 2.4 trillion parameters in total and activates 95 billion per token across 512 experts, arrives as 213 separate files, and handles 262,144 tokens of context natively, extensible to about a million. On Qwen's own scorecard it reaches 86.6 on Terminal Bench 2.1, above Opus 4.8 at 84.6 and below GPT-5.6 Sol at 88.8. The licence is a custom one rather than the Apache 2.0 that Qwen 3.5 and 3.6 shipped under: products with more than 100 million monthly users or $20 million in monthly revenue must display the model's name in their interface, and any model-as-a-service or AI work-assistant business earning more than $50 million over twelve months needs a separate licence from Qwen. The Qwen3.8-27B promised alongside it has not appeared on Qwen's official page. (Correction, 13 August: our original headline called this the largest open-weight model yet. It is not. Moonshot AI released Kimi K3, with 2.8 trillion parameters, on 27 July. Qwen3.8-Max is the largest open-weight model Alibaba has published and the first Max-class Qwen opened, but not the largest overall.)
Related stories
Rumour: the anonymous model that just topped a coding benchmark, for free, is said to be Zhipu's unreleased flagship2026-08-21DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21