Huawei, in collaboration with Beijing Institute of Technology and Peking University, has launched the Claw-Anything benchmark to evaluate AI agents' capabilities in managing digital life tasks. The benchmark, released on May 25, 2026, tests AI models like GPT-5.5 and Claude Opus 4.7 in complex personal assistant roles, with scores of 34.5% and 31.8% respectively. It challenges AI to handle tasks across multiple interdependent services, simulating real-world digital environments. The benchmark involves managing tasks across graphical and command line interfaces, requiring proactive behavior from AI. Huawei's research indicates that specialized training can enhance AI performance, as demonstrated by the Qwen3.5-27B model's 23.7% improvement after fine-tuning. This development is significant for crypto AI projects, which often rely on OpenAI models, suggesting that domain-specific training could enhance their effectiveness in managing complex crypto tasks.