Hermes Agent Overtakes OpenClaw as Top Open-Source AI Agent by Daily Tokens
In a dramatic shift signaling a new era for open-source AI agents, Nous Research's Hermes Agent has seized the top spot on OpenRouter's global daily inference rankings as of May 10, processing a staggering 224 billion tokens against OpenClaw's 186 billion. The upset crowns Hermes as the most-used open-source AI agent, overturning the prior dominance of its rival and highlighting a decisive pivot in developer priorities.
The victory is fundamentally a triumph of engineering philosophy over raw scale. While OpenClaw boasts connectivity to over 50 platforms, a bustling skill marketplace, and more than 44,000 available skills, its growth has been hamstrung by severe security vulnerabilities—including nine disclosed CVEs (Common Vulnerabilities and Exposures) and the identification of over 340 malicious skills circulating in its ecosystem. Compounding these technical risks, leadership uncertainty has plagued the project since its founder joined OpenAI, leaving the community without a clear strategic rudder.
Hermes, by contrast, wins through a tightly focused architecture. It prioritizes self-improving capabilities—most notably, the ability to auto-generate its own skill files—and operates across a deliberate set of 20 core platforms. This streamlined approach not only reduces the attack surface but delivers superior reliability and agility, factors that evidently resonate more deeply with developers than sheer breadth of integration.
The broader implication of this ranking flip is profound. The developer community is clearly voting with their compute cycles for agents that learn from experience rather than those that simply connect to more services. The key differentiator fueling Hermes' ascent is the SKILL.md standardization, a framework that enables skill portability across different agent architectures. By treating skills as modular, version-controlled assets, SKILL.md is accelerating ecosystem maturity and allowing agents like Hermes to compound their improvements far faster than rivals locked into heterogeneous, untrusted skill repositories.
