- Cleus
- AI benchmarks
- FrontierMath
FrontierMathexplained
Unpublished problems from working mathematicians. FrontierMath is hundreds of original maths problems written by professional mathematicians and never published, ranging from hard olympiad level (tier 1) to early research level (tier 3). Answers are exact objects such as a large number or a formula, checked automatically. This page uses Tiers 1–3 of the current version.
- Highest published score
- GPT-6.1 Sol, 93.7%
- Leader among popular models
- GPT-6 Astra
- Models with a result
- 90
- Skill it tests
- Math
What FrontierMath really tests
- What it is
- FrontierMath is hundreds of original maths problems written by professional mathematicians and never published, ranging from hard olympiad level (tier 1) to early research level (tier 3). Answers are exact objects such as a large number or a formula, checked automatically. This page uses Tiers 1–3 of the current version.
- A question like
- Find the exact number of solutions to a particular equation over a finite field, where getting there means combining number theory and algebra the way a researcher would, over many pages of working. Our own example in the style of the test, not a real item.
- Why it matters
- Because the problems are secret, a model can’t have memorised them. It is one of the best signals of real mathematical ability.
- Reading the score
- Share of problems solved with an exactly correct answer.
- The catch
- OpenAI funded the benchmark and has access to many of the problems, apart from a held-out set. Epoch reports this openly.
- Made by
- Epoch AI, 2024–2026. Official page
Leaderboard
Each model’s best published result across its reasoning settings.
Popular models
The models we track across this site.
Top 20 of all 90 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is FrontierMath?
FrontierMath is hundreds of original maths problems written by professional mathematicians and never published, ranging from hard olympiad level (tier 1) to early research level (tier 3). Answers are exact objects such as a large number or a formula, checked automatically. This page uses Tiers 1–3 of the current version.
What does a FrontierMath score mean?
Share of problems solved with an exactly correct answer.
Which AI model is best on FrontierMath?
GPT-6.1 Sol from OpenAI has the highest published FrontierMath result we have (93.7%). Among the popular models on this site, GPT-6 Astra leads with 93.7%.
Is FrontierMath a good benchmark?
Because the problems are secret, a model can’t have memorised them. It is one of the best signals of real mathematical ability. The catch: OpenAI funded the benchmark and has access to many of the problems, apart from a held-out set. Epoch reports this openly.
Who made FrontierMath?
Epoch AI, 2024–2026.
Benchmark scores: Epoch AI, Capabilities & benchmarking, used under CC BY 4.0. For each model we show its best published result across reasoning settings. Prices and context sizes: OpenRouter. Data refreshed November 9, 2026. Hands-on tests (answers, websites, code, refusals, speed) are our own single runs. Explanations written by Cleus.