LLM Leaderboard — compare 338 AI models by capability, price & context

Performance index

26 of 338 models carry a computed score. Models not in the benchmark snapshot show their facts and make no score claim.

Best AI for each capability

How this index is built

Two layers, kept deliberately separate. The factual layer — price per token, context window, modality, launch date — comes from live provider data covering all 338 models, refreshed when the index rebuilds. The capability layer comes from LiveBench, an independent benchmark that publishes dated snapshots and resists contamination by rotating its questions. We compute the LLM Index Score from LiveBench's raw per-task results using weights we publish, and we show every component on every model page.

Nothing here is hand-tuned. No model gets a nudge, no score is adjusted after the fact, and where a third party publishes its own index we display it attributed to them rather than folding it into ours. When a model is missing benchmark data we say so and show no score, instead of guessing one — a partial score is not comparable to a complete one, and an estimate presented as a measurement is worse than no number at all.

Why component transparency matters

Every model leaderboard compresses many abilities into one number, and that compression is where the information goes missing. Two models with the same overall score can be completely different hires: one steady across the board, the other carried by a single very strong category while trailing everywhere else. If the leaderboard only shows you the total, you cannot tell them apart — and you will pick the wrong one for a narrow workload.

So the breakdown is not a footnote here, it is the point. Each model page prints all seven category scores, their weights, their contribution to the total, and the raw task scores underneath. You can add the contributions up yourself and land on the published score. A ranking you can check is worth more than a ranking you have to trust.

FAQ