RouterBench

•  leaderboard  •

The AI Credit Score™
leaderboard.

Live per-task rankings from real traffic — which model wins for code, reasoning, extraction, cost and speed, updated continuously. No vendor bias: every model, including ours, on the same yardstick.

•  snapshot  •

A sample of the board.

Live from real console traffic — success rate, volume and trend per model, refreshed continuously.

live · success rate
1gpt-5-nano
99—
2claude-sonnet-4-5
100—
3gpt-4o-mini
100—
4llama-3.3-70b-instruct
0—

Route on the board, automatically.

You don't have to check the leaderboard — routerbench/auto does, on every request.