OpenRating is an independent, open-source AI rating platform. Join the community

OOpenRating

Open source AI ratings

Independent ratings for every AI model

OpenRating measures intelligence, speed, and price across the major model labs — with open methodology and open data, so you can verify every number yourself.

Live leaderboard

Benchmarks
1

Claude Opus 5 (max)

Anthropic

63.4
2

Claude Fable 5 (with fallback)

Anthropic

62.3
3

GPT-5.6 Sol (max)

OpenAI

61.2
4

Grok 4.6 (high)

xAI

60.8
5

Kimi K3 (max)

Moonshot AI

59.8
View full leaderboard →

OpenRating Index

63.4

Top measured score. One composite metric that weights reasoning, knowledge, and autonomous agentic ability.

Cheapest frontier-class

$0.3

Input price per million tokens for frontier models scoring 55+ on the OpenRating Index.

Ratings covering the major model labs

OpenAIAnthropicGoogleMetaxAIDeepSeekMistralCohere

Methodology

Three axes. One transparent method.

Every rating is computed from public benchmark runs and measured API performance. Raw data and methodology are open source.

OpenRating Index

A weighted composite of reasoning, knowledge, and agentic benchmarks, normalised to a 0–100 scale.

See the leaderboard

Speed Index

Blended output speed measured in tokens per second under standardised load, plus time-to-first-token.

Sort by speed

Task Economics & Price

Empirical cost per standard benchmark task, plus list input and output prices per million tokens.

Inspect cost breakdown

Independent by design

No vendor money. No black boxes.

  • Open data

    Every benchmark result is published as raw data you can download and re-run.

  • Open methodology

    The rating formula is public and versioned. Anyone can audit or improve it.

  • Independent

    No lab sponsorship. Ratings are computed by the community, for the community.

Ratings covering the major model labs

OpenAIAnthropicGoogleMetaxAIDeepSeekMistralCohere

Put two models side by side

Compare intelligence, speed, price, and context windows across any combination of models.

Benchmark Newsletter

Get new benchmark ratings in your inbox

Monthly empirical updates when labs drop new models. No marketing spam.

Join 4,200+ ML engineers and researchers. Unsubscribe at any time.