Always-on community report
reported 62.3M tokens per month, taken as a 30-day figure
cited source (2026-07-21) ↗Ranked by 30-day gross tokens. First-party-attested exporter runs only — community self-reports live in their own field below and are never ranked here.
How much of the gross volume was served from prompt cache instead of fresh computation.
11.54B cache-read + 0 cache-write of 12.13B gross. Reasoning tokens tracked separately: 6.1M (never double-counted).
Honesty clause: cache reads are re-served context, priced and computed differently from fresh tokens. This site never equates cache tokens with billable tokens — that is why fresh volume is always shown alongside gross. Formulas on methodology.
Per-UTC-day aggregates for Herman, window ending 2026-07-30. Usage is attributed to the day its accounting row was last updated — see limitations.
| date (UTC) | gross tokens | fresh tokens | api calls |
|---|---|---|---|
| 2026-06-30 | 632,188,870 | 19,403,905 | 2,669 |
| 2026-07-01 | 1,082,632,204 | 53,088,078 | 4,254 |
| 2026-07-02 | 371,759,919 | 20,210,420 | 1,715 |
| 2026-07-03 | 561,722,517 | 26,834,394 | 2,141 |
| 2026-07-04 | 435,120,433 | 33,765,658 | 1,848 |
| 2026-07-05 | 250,108,594 | 6,478,316 | 1,083 |
| 2026-07-06 | 970,021,263 | 33,829,197 | 3,437 |
| 2026-07-07 | 1,148,667,316 | 37,254,059 | 4,238 |
| 2026-07-08 | 1,313,225,887 | 34,537,744 | 3,968 |
| 2026-07-09 | 1,073,347,494 | 44,005,377 | 3,283 |
| 2026-07-10 | 595,183,362 | 17,748,751 | 2,157 |
| 2026-07-26 | 741,076,225 | 43,252,571 | 6,079 |
| 2026-07-27 | 1,005,497,395 | 67,436,416 | 9,385 |
| 2026-07-28 | 863,640,605 | 58,139,368 | 8,311 |
| 2026-07-29 | 548,289,417 | 48,624,078 | 5,601 |
| 2026-07-30 | 532,700,728 | 40,277,692 | 5,478 |
Anonymous public self-reports, quoted for scale context. Unverified. Not ranked. Not comparable to attested entries. Full framing on compare.
reported 62.3M tokens per month, taken as a 30-day figure
cited source (2026-07-21) ↗reported approximately 1B tokens per month, taken as a 30-day figure
cited source (2026-07-21) ↗reported approximately 100M tokens per day; shown as a 30-day extrapolation, clearly labeled
cited source (2026-07-21) ↗One point-in-time, app-wide aggregate — every Hermes Agent instance routed through OpenRouter combined. Not one instance; excluded from all rankings.
One stdlib-only script, read-only against your own database, strict allowlist out. No accounts, no telemetry, no identity.
$ python3 scripts/export_anonymous.py \
--db "$HOME/.hermes/state.db" --days 30 --alias YourAlias \
--instance-secret-file "$HOME/.config/tokenmaxxing/secret" --create-secret \
--out my-snapshot.json
$ python3 scripts/validate_snapshots.py my-snapshot.json
OK my-snapshot.json