Qwen3.8-Max arrives with a bold claim: it outperforms GPT-5.6 Sol Max and Fable 5 on agentic computer use

Alibaba’s Qwen team introduced Qwen3.8-Max, a 2.4T-parameter mixture-of-experts multimodal model aimed at autonomous software engineering and enterprise agent tasks. The company claims it leads on OSWorld-Verified and PaperBench while staying competitive across coding, research, multimodal reasoning, and visual web development benchmarks. Alibaba also says open weights are coming next week, though licensing terms remain undisclosed.

If independently validated, Qwen3.8-Max would reinforce the shift from general chatbot benchmarks toward agentic computing as a key frontier AI battleground. An open-weight release could accelerate enterprise self-hosting and pressure proprietary model providers, but the impact will depend heavily on whether Alibaba uses a permissive or restrictive license.