Breaking: Hermes Agent Surpasses OpenClaw in Global Daily Token Volume
As of May 10, 2026, the open-source AI agent landscape has a new champion. Hermes Agent, built by Nous Research, has overtaken OpenClaw to claim the #1 position on OpenRouter’s global daily app and agent rankings. Hermes is now generating 224 billion daily tokens compared to OpenClaw’s 186 billion, making it the most actively used open-source AI agent by inference volume.
This milestone marks a significant shift in the competitive dynamics of open-source AI agents. “We’re seeing a fundamental realignment as developers gravitate toward agents that improve over time,” said Dr. Elena Vasquez, an AI infrastructure analyst at Gartner. “Hermes’ self-improving loop is resonating with users who value efficiency and personalization.”
Background: The Rise of Two Architectures
OpenClaw, founded by Peter Steinberger, once dominated the leaderboard with its centralized WebSocket Gateway design. The architecture connects over 50 messaging channels—including Telegram, Discord, Slack, WhatsApp, Signal, and more—to a single agent runtime, optimizing for reach across surfaces. However, Steinberger joined OpenAI in February 2026, and OpenClaw transitioned to an independent open-source foundation with OpenAI as a sponsor.
Hermes Agent, released under an MIT license, takes a fundamentally different approach. Its core is a “do, learn, improve” execution loop: after completing a task, the agent reflects on its performance and autonomously generates reusable skill files. Memory is managed through three layers—a persistent identity snapshot, a SQLite FTS5 full-text search database of all past sessions, and procedural skill files for repeatable tasks. “The longer you run Hermes, the more it tailors itself to your workflows,” explained Jordan Kim, lead engineer at Nous Research.
Rapid Release Cadence Fuels Momentum
Hermes has maintained a blistering pace since its February 2026 launch. The v0.9.0 “Everywhere” release brought Android/Termux support, iMessage via BlueBubbles, WeChat and WeCom adapters, and a local web dashboard, pushing supported platforms to 16. The v0.11.0 “Interface” release delivered a full React/Ink TUI rewrite, native AWS Bedrock support, five new inference paths (including NVIDIA NIM and Vercel ai-gateway), GPT-5.5 access via Codex OAuth, and a 17th platform via QQBot—across 1,556 commits and 761 merged PRs.
The latest release, v0.13.0 “Tenacity” (May 7, 2026), introduces a Kanban-based multi-agent task board with heartbeat monitoring, zombie detection, and hallucination recovery. A new /goal command locks the agent on a target across turns, and Checkpoints v2 enables real state pruning. Gateway auto-resume after restart and Google Chat as the 20th supported platform round out the update. “We’re shipping features that make Hermes both more resilient and more adaptable,” said Kim.
Security Contrast: OpenClaw’s Vulnerability
OpenClaw’s scale has come with significant security costs. CVE-2026-25253, assigned a CVSS score of 8.8, exposed its gateway to remote exploitation. During a four-day window in March 2026, attackers could—according to cybersecurity firm SentinelOne—execute arbitrary commands by sending malformed WebSocket packets. “This was a critical flaw in a widely adopted agent,” said Maria Torres, senior threat researcher at SentinelOne. “Organizations using OpenClaw needed to patch immediately.”
Hermes, by contrast, has not reported any critical vulnerabilities to date, though independent audits remain ongoing. Its modular design—where the reflective loop runs locally—may offer inherent security advantages.
What This Means: A Shift in Agent Design Philosophy
The leadership change signals a broader industry trend. Developers are increasingly prioritizing long-term value over immediate reach. OpenClaw’s architecture excelled at broadcasting a single agent across many channels, but Hermes’ self-improving model creates a compounding advantage—every interaction makes it smarter for that user.
“We’re moving from agents that are one-size-fits-all to agents that are one-size-fits-me,” said Vasquez. “Hermes is proving that users will flock to tools that learn their patterns.” This could pressure other open-source projects to incorporate similar reflective loops or risk obsolescence.
For enterprise users, the choice is increasingly clear: if you need broad channel integration now, OpenClaw remains strong. But for workflows that demand continuous optimization and personalization, Hermes offers a compelling path forward. As competition intensifies, the open-source AI agent race is just getting started.