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

Alibaba's Qwen3.8-Max just claimed the crown for agentic computer use, beating GPT-5.6 Sol Max on key benchmarks. This 2.4T model is built for autonomous software engineering and long-horizon tasks, not just chatting.

For builders, this shifts the goalposts from 'can it talk' to 'can it actually do the job.' If the upcoming open-weights release is permissive, it breaks the US monopoly on frontier-class agentic models.

Alibaba is weaponizing low pricing and high performance to squeeze US labs, forcing a shift from conversational AI to autonomous software agents.

• →Build 'agent-first' tools for legacy software that lacks APIs by using the model's computer-use abilities.
• →Deploy self-hosted, Max-class models for high-security R&D using the upcoming open-weights release.
• →Target mid-market enterprises with high-volume automation tasks that are too expensive to run on Claude or GPT.

• →The 'open weights' release might come with restrictive Chinese licensing that prevents commercial use in Western markets.
• →Alibaba's claims about 10-day software projects are unverified and might not translate to real-world reliability.

[{"heading":"From Chatbots to Coworkers","body":"The industry is moving past the chat era. Qwen is betting that the real value is in models that stay active for days to finish complex projects rather than just answering prompts."},{"heading":"The Open-Weight Wildcard","body":"The open-weights announcement is the real wildcard. If Alibaba releases a Max-class model under a permissive license, it breaks US dominance. But if the license is restrictive, the hype dies fast."},{"heading":"The Economics of Intelligence","body":"By pricing Qwen significantly below GPT-5.6 and Claude, Alibaba is turning frontier intelligence into a commodity. This forces US labs to defend their margins or lose the scale game."},{"heading":"What to Watch","body":"Watch the license terms next week. Also, look for independent verification of the OSWorld scores. If these numbers hold up, the agentic wars are officially on."}]