Alibaba's Qwen3.7-Max runs 35-hour autonomous agent task

Alibaba's Qwen3.7-Max has captured the AI community's attention with a demonstration of unprecedented autonomous endurance. The model successfully executed a 35-hour kernel optimization task that required over 1,000 tool calls—all without any human intervention. Notably, it operated on unfamiliar hardware, showcasing adaptability beyond its training environment. This feat moves the goalpost from single-task benchmarks to sustained, long-horizon autonomy, a critical capability for real-world AI agents that must function independently over extended periods.

The model's 10x inference speedup further amplifies its impact, making it not just autonomous but also efficient. Qwen3.7-Max's performance highlights Alibaba's competitive edge in the agentic AI race, where the ability to plan, execute, and recover over hours—rather than minutes—is becoming the new standard. As discussions swirl around agent reliability and scalability, this demonstration suggests that foundation models are increasingly capable of handling complex, multi-step workflows without constant human oversight, paving the way for more autonomous enterprise solutions.