Alibaba Qwen3.8-Max: AI Codes 10 Days Autonomously

Alibaba has unveiled its new flagship AI model, Qwen3.8-Max, which autonomously coded a complete software project for over ten days without human intervention. This model wrote its own code, ran tests, and merged pull requests. What's interesting is that Qwen3.8-Max uses 2.4 trillion parameters but activates only 95 billion at a time. It immediately became the highest-ranking Chinese model for text tasks on Arena.AI, and second globally for vision tasks. The model's pricing is set at $2 per million input tokens and $6 per million output, making it significantly cheaper than some competitors. Beyond coding, it also optimized a cryptographic hardware accelerator, reducing its design from 8,298 logic gates to 678. Alibaba plans to open-source the full weights for Qwen3.8-Max within a week, a change from its previous strategy of keeping top-tier models proprietary. This could be a significant development for enterprise adoption. The bottom line is that this model shows impressive autonomous capabilities, potentially impacting how software and hardware are developed.