Hermes Agent overtakes OpenClaw in global AI agent usage race

Tuesday 12 May 2026 - 13:27

By: Dakir Madiha

Hermes Agent overtakes OpenClaw in global AI agent usage race

Hermes Agent from Nous Research has become the most widely used open source AI agent on OpenRouter, surpassing longtime leader OpenClaw after a rapid surge in developer adoption and token usage.

Public application rankings on OpenRouter showed on May 10 that Hermes Agent was processing roughly 224 billion tokens per day, compared with 186 billion for OpenClaw. The milestone capped an aggressive rise for the framework, which launched on February 25, 2026 and briefly peaked at 271 billion daily tokens earlier this month. The shift signals a broader realignment in the open source AI agent ecosystem, where developers are increasingly prioritizing reliability, persistent memory systems and flexible model compatibility.

OpenClaw’s decline accelerated after a series of operational and security setbacks during the first quarter of 2026. Its creator, Peter Steinberger, announced in February that he would join OpenAI while OpenClaw transitioned toward a foundation structure. At the same time, the platform faced mounting criticism over multiple vulnerabilities, including command injection flaws, remote code execution exploits and privilege escalation bugs identified by cybersecurity firms. Security researchers also reported the spread of hundreds of malicious skills through ClawHub, exposing users to infostealer malware campaigns targeting developer systems.

The platform suffered another setback in April after Anthropic revoked OAuth subscription tokens that many OpenClaw users relied on to access Claude models at reduced cost. Developers reported a sharp drop in activity almost immediately after the change, weakening OpenClaw’s position across the broader agent ecosystem. Traffic and token usage fell as developers shifted toward alternatives with more stable infrastructure and fewer dependency risks.

Hermes Agent gained momentum by introducing a self improving learning loop that preserves knowledge between sessions instead of restarting every interaction from scratch. The framework extracts reusable skills from completed tasks and automatically reloads them when similar requests appear again. Its architecture also supports multiple providers, allowing developers to connect through OpenRouter, OpenAI APIs or other compatible endpoints without locking themselves into a single model ecosystem.

The framework’s rapid release cycle also strengthened community support. Hermes reached version 0.8.0 by early April after more than 1,000 pull requests had been merged. Developers on technical forums increasingly described the platform as moving beyond simple session based assistants toward autonomous AI operators capable of long term adaptation and memory retention.

Despite its rapid growth, Hermes now faces the same challenges that weakened OpenClaw, particularly around security, scalability and governance. As adoption increases, researchers and developers will closely monitor whether the platform can maintain trust while managing a growing ecosystem of third party skills and integrations.