AI Model Leaderboard
Live latency, cost, and quality rankings across 0 models. Latency is measured from real production traffic; quality is scored against open-source benchmarks (HumanEval, GSM8K, MMLU subsets).
Top quality
—
no benchmarks yet
Fastest
—
no traffic data
Cheapest
—
no pricing
| Rank | Model | Provider | Quality | Latency p50 | Latency p95 | Input $/1M | Output $/1M | Context |
|---|---|---|---|---|---|---|---|---|
| No model data available yet. Daily benchmark runs will populate this within 24 hours. | ||||||||
Methodology
- Latency — p50 / p95 wall-clock from real production traffic (last 7 days, successful requests only). Includes our gateway hop, so it reflects what your users would actually experience.
- Cost — input + output prices per 1M tokens, sourced directly from each provider’s public pricing page. Updated when providers change pricing.
- Quality — 0.0–1.0 score on a curated suite of HumanEval (code), GSM8K (math), and MMLU (reasoning) prompts. Each prompt has either an exact-match, numeric, substring, or embedding-similarity grader. Run daily on every active model.
- No quality score — the model hasn’t been benchmarked yet. New models hit the leaderboard within 24 hours of their first request.
Spotted a model that’s missing or wrong? Tell us.
Use any of these models with one API key.
Get an API key — free $5 credit