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

Alibaba has officially released Qwen3.8-Max, a massive flagship model featuring 2.4 trillion parameters designed to set a new standard for autonomous coding and complex workplace collaboration. This marks a monumental milestone as Alibaba plans to open-source the weights of this Max-class model, offering frontier-level intelligence with local hardware independence. [1, 2, 3]

Core Technical Specifications

Parameter Scale: Scales to 2.4 trillion total parameters with 95 billion active parameters using a sparse Mixture of Experts (MoE) architecture.
Context Window: Features a massive 1 million-token context window to effortlessly handle entire software repositories.
Affordable Access: Priced via QwenCloud at $2 input / $6 output per million tokens, significantly undercutting Western frontier models.
Top Performance: Debuted at #4 on the Frontend Code Arena leaderboard, matching elite proprietary models like Claude Opus 5. [1, 2, 3, 4, 5]

Key Capabilities & Breakthroughs

10+ Day Autonomous Coding
The model can build, maintain, and upgrade software entirely on its own. In live tests tracked on the oh-my-cli GitHub repository, Qwen3.8-Max operated fully autonomously for over 16 days—managing issues, writing code, executing self-testing loops, and merging 127 production-ready Pull Requests without human oversight. [1]
Long-Horizon Strategy & Crisis Management
Built for system-level planning, it can execute over 500 turns of chip design optimization or manage a 365-day e-commerce operational strategy. In real-world simulations, it adaptive-learned under heavy financial pressure, outpacing competing models like GLM 5.2 to achieve a massive 4.16x capital investment return. [1, 6]
Native Multimodal Loops
Unlike models that use vision just as an input layer, Qwen3.8-Max uses vision as a continuous visual feedback loop. It looks at its own generated web layouts, UI animations, or desktop environments, spots rendering bugs, and self-corrects the code in real-time. [6, 7, 8]

Availability & Deployment

The preview version is currently live and testable via the QwenCloud API and developer tools like Qoder AI. For local developers and businesses wanting complete privacy, the official open-weight release is scheduled to drop next week. [1, 3, 8]

Would you like help setting up a QwenCloud API connection, exploring the autonomous GitHub loop architecture, or planning your local hardware specs for when the weights drop next week?