Huawei's New Benchmark Gives AI Agents Months of Your Life—Then Watches Them Fail

Claw-Anything simulates a real digital existence and asks AI assistants to handle it. GPT-5.5, the best model available, scored 34.5%.