AI Agents on OpenRouter
Real-world adoption ranked by token volume on OpenRouter — the neutral marketplace for LLM inference. Because vendors don't benefit from gaming OpenRouter numbers (it's pay-per-token), this is one of the few adoption signals that's resistant to benchmark-style gaming.
Auto-updated daily·Last snapshot: Jun 23, 2026·Δ column = change vs ~7 days prior
| # | Agent | Tokens / week | Δ 7d | 28d | Cumulative | Updated |
|---|---|---|---|---|---|---|
| 01 | Hermes Agent Nous ResearchOSS | 1.03T | +22% | 7.72T | Jun 23, 2026 | |
| 02 | Cline ClineOSS | 444.4B | 0.0% | 4.54T | Jun 17, 2026stale | |
| 03 | OpenClaw Erik SteinbergerOSS | 184B | +3.4% | 9.72T | Jun 23, 2026 | |
| 04 | Claude Code Anthropic | 142B | +1.4% | 3.52T | Jun 23, 2026 |
13other tracked agents don't use OpenRouter (most route directly to their own provider — Claude Code → Anthropic, Codex → OpenAI, Cursor → multi-provider via own infra) so they're excluded from this ranking.
How this works
OpenRouter publishes a public "Top apps" leaderboard showing weekly token volume per AI agent — the most gaming-resistant adoption signal we know of (vendors don't benefit from inflating numbers; it's pay-per-token).
Numbers come from openrouter.ai/rankings. A headless-browser scraper runs nightly at 06:30 UTC on the VPS — it loads the rankings page in real Chromium, reads the rendered DOM, and writes a fresh row per agent into our snapshots table. Plausibility checks reject any 5×-or-more jump from the prior reading.
The Δ 7d column compares the latest snapshot to the closest one ~7 days back. The 28-day sparkline reflects 4 weeks of history. See the methodology page for ingestion details.
Other leaderboards
- gpt.buzz Composite Agent Index — cross-benchmark capability ranking