About & docs

How ModelArena works, where the data comes from, and how to use it.

What this is

ModelArena is a live leaderboard of LLM models. It tracks 339 models (108 with Artificial Analysis benchmark indexes), their pricing, context windows, and design-arena ELO ratings — and records a dated snapshot every day so you can watch how models move over time. Currently tracking 338 models as of 2026-08-05.

Data sources

  • OpenRouter /api/v1/models (primary, auto) — model catalog, pricing, context, and each model's Artificial Analysis indexes (Intelligence, Coding, Agentic) plus Design Arena ELO. Refetched automatically once a day by a scheduled Netlify function.
  • Curated classic evals (secondary) — MMLU-Pro, GPQA, HumanEval, MATH/AIME etc. seeded from public model cards.
  • Runtime — Next.js + MongoDB Atlas + Netlify. Source at git.

How fresh is it?

A snapshot runs once a day (06:00 UTC) via a Netlify scheduled function. You can also trigger an immediate refresh yourself with the manual sync endpoint using the site's sync key.

Caveats

The Artificial Analysis indexes are third-party composites — they are estimates, not ground truth. Classic evals are manually curated and may lag model releases. Benchmark deltas between snapshots should be read as relative movement,not winsorized differences.

Pages