Alibaba Releases Qwen3.8-Max: 2.4T Parameters, Open Weights Next Week
AGI Hunt · wechat · 2026-08-03
Alibaba has released Qwen3.8-Max, featuring 2.4 trillion total parameters (95B active), 1M context length, and native vision capabilities. It scores 1668 on FrontendCodeArena, just 1 point behind Claude Opus 5 High.
Key Highlights:
- Open Source: Weights arriving next week, marking the first open-source Max-tier model, alongside a 27B variant
- Pricing: $2/M input, $6/M output tokens — output cost is only 1/5 of GPT-5.6 Sol
- Autonomous Capabilities: Independently managed projects (265 commits in 16 days), reproduced and improved research papers (AIME24 +2.7 over original), competed in data science contests (beating 87% of human teams), and optimized chip logic (81% area reduction)
- Gaps: Trails Fable5 on SWE-bench Pro and GPT-5.6 Sol on TerminalBench 2.1
Community attention is focused on whether the license will be MIT.
More from coding & agent
- Can AI Coding Agents Crack Nvidia's CUDA Moat? — bingxu_ · 2026-08-04
- HeyGen LiveAvatar Agent Builds $3M Pipeline During 8-Week Employee Leave — HeyGen · 2026-08-04
- Agent Benchmark Reflections: Scores Are Deceptive, Open-source Trajectories Needed — Shahules786 · 2026-08-04
- Agent Benchmark Flaws: Over-specified Verifiers Penalize Semantically Correct Actions — Shahules786 · 2026-08-04
- ITSMBench: Open-Sourcing a Benchmark for Enterprise AI Agents — Shahules786 · 2026-08-04
- Protecting AI Attention: The Essence of Inference Efficiency — DanWahlin · 2026-08-04