evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

FrontierMath

A frontier math benchmark with constrained public access and source-linked result claims.

ReasoningaccuracyHigher is better
ScoresAboutRun this benchmark

What is FrontierMath?

A frontier math benchmark with constrained public access and source-linked result claims. evals.report tracks reported FrontierMath scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported FrontierMath score: GPT-5.5 Pro — 52.4% (accuracy).

ModelLabScore↓Source modelStatusDate
GPT-5.5 ProOpenAI52.4%GPT-5.5 ProOfficialApr 23, 2026Details
GPT-5.5OpenAI51.7%GPT-5.5OfficialApr 23, 2026Details
GPT-5.4 ProOpenAI50.0%GPT-5.4 ProOfficialMar 5, 2026Details
GPT-5.4OpenAI47.6%GPT-5.4OfficialMar 5, 2026Details
Claude Opus 4.7Anthropic43.79%Claude Opus 4.7OfficialApr 16, 2026Details
Claude Opus 4.6Anthropic40.7%Claude Opus 4.6OfficialFeb 5, 2026Details
GPT-5.2OpenAI40.7%GPT-5.2OfficialDec 11, 2025Details
Muse SparkMeta39.0%Muse SparkOfficialApr 8, 2026Details
Gemini 3.5 FlashGoogle DeepMind38.97%Gemini 3.5 FlashOfficialMay 19, 2026Details
Kimi K2.6OpenMoonshot AI38.97%Kimi K2.6OfficialApr 20, 2026Details
Gemini 3 ProGoogle DeepMind37.6%Gemini 3 ProOfficialNov 18, 2025Details
Gemini 3.1 Pro PreviewGoogle DeepMind36.9%Gemini 3.1 ProOfficialFeb 19, 2026Details
Gemini 3 FlashGoogle DeepMind35.64%Gemini 3 FlashOfficialDec 17, 2025Details
GLM-5.1OpenZ.ai33.45%GLM-5.1OfficialApr 7, 2026Details
GPT-5OpenAI32.41%GPT-5OfficialAug 7, 2025Details
Claude Sonnet 4.6Anthropic32.4%Claude Sonnet 4.6OfficialFeb 17, 2026Details
GPT-5.1OpenAI31.03%GPT-5.1OfficialNov 12, 2025Details
Kimi K2.5OpenMoonshot AI27.9%Kimi K2.5OfficialJan 27, 2026Details
GPT-5 miniOpenAI27.24%GPT-5 miniOfficialAug 7, 2025Details
Qwen 3.6 PlusAlibaba / Qwen26.21%Qwen 3.6 PlusOfficialApr 2, 2026Details
o4-miniOpenAI24.83%o4-miniOfficialApr 16, 2025Details
Qwen 3.6 Max PreviewAlibaba / Qwen23.1%Qwen 3.6 Max (Preview)OfficialApr 20, 2026Details
DeepSeek V3.2OpenDeepSeek22.1%DeepSeek-V3.2OfficialDec 1, 2025Details
Claude Opus 4.5Anthropic20.69%Claude Opus 4.5OfficialNov 24, 2025Details
Grok 4xAI19.66%Grok 4OfficialJul 9, 2025Details
o3OpenAI18.69%o3OfficialApr 16, 2025Details
GLM-5OpenZ.ai16.43%GLM-5OfficialFeb 11, 2026Details
Claude Sonnet 4.5Anthropic15.22%Claude Sonnet 4.5OfficialSep 29, 2025Details
Gemini 2.5 ProGoogle DeepMind14.14%Gemini 2.5 Pro (Jun 2025)OfficialMar 25, 2025Details
Qwen3 235B A22B Instruct 2507OpenAlibaba / Qwen8.48%Qwen3-235B-A22B (Jul 2025)OfficialJul 21, 2025Details
Claude Opus 4.1Anthropic7.24%Claude Opus 4.1OfficialAug 5, 2025Details
Claude Haiku 4.5Anthropic5.9%Claude Haiku 4.5OfficialOct 15, 2025Details
GPT-4.1OpenAI5.52%GPT-4.1OfficialApr 14, 2025Details
Gemini 2.5 FlashGoogle DeepMind4.84%Gemini 2.5 FlashOfficialApr 17, 2025Details
Claude Opus 4Anthropic4.48%Claude Opus 4OfficialMay 22, 2025Details
Claude Sonnet 4Anthropic4.14%Claude Sonnet 4OfficialMay 22, 2025Details
Claude 3.7 SonnetAnthropic4.14%Claude 3.7 SonnetOfficialFeb 24, 2025Details
GLM-4.6OpenZ.ai3.82%GLM-4.6OfficialSep 30, 2025Details
Gemini 2.0 FlashGoogle DeepMind1.72%Gemini 2.0 FlashOfficialDec 11, 2024Details
DeepSeek V3OpenDeepSeek1.72%DeepSeek-V3OfficialDec 26, 2024Details
Claude 3.5 SonnetAnthropic1.03%Claude 3.5 SonnetOfficialJun 20, 2024Details
Llama 4 MaverickOpenMeta0.69%Llama 4 MaverickOfficialApr 5, 2025Details
GPT-4oOpenAI0.34%GPT-4oOfficialMay 13, 2024Details
Llama 4 ScoutOpenMeta0.0%Llama 4 ScoutOfficialApr 5, 2025Details

Each row reports the model’s accuracy on FrontierMath. Click a row for the full run context.