Alibaba Qwen3.7-Max optimizes chip code autonomously for 35 hours

alibaba's qwen team launched qwen3.7-max, a proprietary model designed exclusively for agent-based workflows via api. in a rigorous test, the model performed 1,158 tool calls over 35 hours to optimize sglang inference code for custom t-head accelerators, resulting in a 10x speedup compared to a 7.3x gain by glm 5.1. additionally, the model functioned as a training watchdog, detecting reward hacking and writing 13 new validation rules to ensure data integrity.