Qwen3.8-Max: A New Bar for Coding and Cowork

Qwen Team released Qwen3.8-Max, a 2.4-trillion-parameter model now on QwenCloud, with open weights coming next week.

• First Qwen-Max-class model with open weights, releasing next week on Hugging Face and ModelScope.
• 2.4T parameters (95B active), available via QwenCloud API with reasoning_effort levels xhigh, medium, and low.
• In autonomous tests, reproduced a research paper and beat its method by +2.7 points on AIME24; beat 458 of 526 human teams in a 24-hour contest.
• Achieved ¥416,252 total balance (4.16x return) in E-Commerce Bench, 38% above second-place GLM 5.2.
• Integrates with Claude Code, Codex, Qoder CLI, Qwen Code, and OpenClaw via API.

Developers and enterprises can now use a frontier-scale model via API and will soon be able to self-host it, since Qwen3.8-Max is the first Max-class model with open weights.

Sources

• Qwen: Qwen3.8-Max: A New Bar for Coding and Cowork

Read this as text

Back to the AI news