All AI performance rankings
AI rankings for math and science
Which AI is best at math and science?
As of Nov 2026: #1 Claude Fable 5.2, #2
Gemini 4 Carbon, #3
Gemini 4 Neon. A math and science performance ranking of 135 AI models across 15 evaluations, including Humanity's Last Exam · CritPt · Arena Math.
Last updated 2026-11-21 · AIB compiles and analyzes scores from many benchmarks to produce its own independent ranking.
This ranking combines the results of 15 tests for this use. Featured tests, recently updated tests and tests that many companies take part in count for more. Only models with scores on 2 or more tests are ranked. Scores are scaled so that the average model is 100; higher is better.
- 🥇AnthropicClaude Fable 5.28 tests counted134.1—35.7%1507
- 🥈GoogleGemini 4 Carbon6 tests counted132.5—36.0%—
- 🥉GoogleGemini 4 Neon6 tests counted131.3——1535
- #4OpenAIGPT-6.2 Astra4 tests counted129.8———
- #5AnthropicClaude Opus 5.510 tests counted129.661.4%31.7%1517
- #6OpenAIGPT-6.1 Sol10 tests counted127.352.9%31.7%1499
- #7AnthropicClaude Fable 5.111 tests counted127.259.1%31.1%1530
- #8OpenAIGPT-6 Astra11 tests counted126.754.7%31.7%1477
- #9GoogleGemini 4 Argon9 tests counted123.957.1%27.1%1535
- #10AnthropicClaude Sonnet 5.510 tests counted123.555.0%31.4%1497
- These models have scores on fewer than 2 tests, so they are left out of the ranking. They join it once they have 2 or more.
Recent ranking changes
Frequently asked questions
- How is the ranking decided?
- Models are ordered by a combined score built from the results of 15 tests for this use. Featured tests, recently updated tests and tests that many companies take part in count for more, and a model with scores on only a few tests is pulled toward the average of models released around the same time. Only models with scores on 2 or more tests are ranked. Scores are scaled so that the average model is 100; higher is better. Pick one test above the table (Humanity's Last Exam, CritPt, Arena Math) to reorder by that test.
- Where do the scores come from?
- We take the scores each test publishes, as they are. This ranking covers 15 tests and 135 AI models, and scores the AI maker reported itself are marked with a dotted underline.
- How often does the ranking change?
- We check test results every day and recalculate the ranking every hour. Results were last collected on 2026-11-21. Past changes are listed under Recent ranking changes.
- Which subscriptions include the top-ranked models?
- The plan comparison shows which models each ChatGPT, Claude, Gemini, Grok and other plan includes, with monthly prices and usage limits taken from the official pricing pages.
- What if a service is slow or down?
- The status page shows whether ChatGPT, Claude, Gemini, Grok and others are working right now and how many outages they had recently, based on their official status pages.