Leaderboard — 24h avg, tokens/sec
| # | model | provider | samples | avg tps | min tps | max tps | avg ttft | last | spark |
|---|---|---|---|---|---|---|---|---|---|
| loading… | |||||||||
Recent probes
last 50| when | model | provider | tps | ttft |
|---|---|---|---|---|
| loading… | ||||
Recent failures
last 50| when | model | stage | error |
|---|---|---|---|
| loading… | |||
About this data
Every 15 minutes we fire a short probe completion at each tracked model on
OpenRouter,
then call GET /api/v1/generation to fetch the authoritative measured stats
(tokens_per_second, time_to_first_token, latency,
provider_name). Snapshots are persisted to SQLite.
No mocks. No Math.random() jitter. If OpenRouter is unreachable we record a
failure row and skip — the UI never shows fabricated numbers.
Embed a per-model badge in your README:
https://holyai.me/tps-board/api/badge?slug=meta-llama/llama-3.3-70b-instruct:free&metric=tps