evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

GPQA Diamond

A difficult subset of GPQA for graduate-level science question answering evaluation.

ReasoningaccuracyHigher is better
ScoresAboutRun this benchmark

What is GPQA Diamond?

A difficult subset of GPQA for graduate-level science question answering evaluation. evals.report tracks reported GPQA Diamond scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported GPQA Diamond score: Fugu — 95.5% (accuracy).

ModelLabScore↓Source modelStatusDate
FuguSakana AI95.5%FuguVerifiedJun 15, 2026Details
Fugu UltraSakana AI95.5%Fugu UltraVerifiedJun 15, 2026Details
GPT-5.4 ProOpenAI94.6%GPT-5.4 ProOfficialMar 5, 2026Details
Claude Mythos PreviewAnthropic94.6%Claude Mythos PreviewUnverifiedApr 7, 2026Details
GPT-5.6 SolOpenAI94.6%GPT-5.6 SolVerifiedJul 9, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind94.1%Gemini 3.1 ProOfficialFeb 19, 2026Details
GPT-5.5OpenAI94.0%GPT-5.5OfficialApr 23, 2026Details
GPT-5.5 ProOpenAI93.9%GPT-5.5 ProOfficialApr 23, 2026Details
Claude Opus 4.8Anthropic93.6%Claude Opus 4.8VerifiedMay 28, 2026Details
GPT-5.4OpenAI93.3%GPT-5.4OfficialMar 5, 2026Details
GPT-5.6 TerraOpenAI92.9%GPT-5.6 TerraVerifiedJul 9, 2026Details
Gemini 3.5 FlashGoogle DeepMind92.8%Gemini 3.5 FlashOfficialMay 19, 2026Details
Gemini 3 ProGoogle DeepMind92.6%Gemini 3 ProOfficialNov 18, 2025Details
GPT-5.6 LunaOpenAI92.3%GPT-5.6 LunaVerifiedJul 9, 2026Details
GPT-5.2OpenAI91.4%GPT-5.2OfficialDec 11, 2025Details
GLM-5.2OpenZ.ai91.2%GLM-5.2VerifiedJun 16, 2026Details
Kimi K2.6OpenMoonshot AI90.8%Kimi K2.6OfficialApr 20, 2026Details
Nex-N2-ProOpenNex AGI90.7%Nex-N2-ProVerifiedJun 2, 2026Details
Claude Opus 4.6Anthropic90.5%Claude Opus 4.6OfficialFeb 5, 2026Details
Claude Opus 4.7Anthropic90.2%Claude Opus 4.7OfficialApr 16, 2026Details
DeepSeek V4 ProOpenDeepSeek90.1%DeepSeek V4 Pro MaxUnverifiedApr 24, 2026Details
Muse SparkMeta89.8%Muse SparkOfficialApr 8, 2026Details
Qwen 3.6 Max PreviewAlibaba / Qwen89.1%Qwen 3.6 Max (Preview)OfficialApr 20, 2026Details
Doubao Seed 2.0 ProByteDance88.9%Seed2.0 Pro (0215)VerifiedFeb 14, 2026Details
Qwen3.5-397B-A17BOpenAlibaba / Qwen88.4%Qwen3.5-397B-A17BUnverifiedFeb 16, 2026Details
DeepSeek V4 FlashOpenDeepSeek88.1%DeepSeek-V4-Flash (Think-Max)VerifiedApr 24, 2026Details
GLM-5OpenZ.ai87.8%GLM-5OfficialFeb 11, 2026Details
GPT-5.1OpenAI87.6%GPT-5.1OfficialNov 12, 2025Details
Kimi K2.5OpenMoonshot AI87.6%Kimi K2.5OfficialJan 27, 2026Details
Qwen 3.6 PlusAlibaba / Qwen87.4%Qwen 3.6 PlusOfficialApr 2, 2026Details
Claude Sonnet 4.6Anthropic87.4%Claude Sonnet 4.6OfficialFeb 17, 2026Details
InklingOpenThinking Machines Lab87.2%InklingVerifiedJul 15, 2026Details
Grok 4xAI87.0%Grok 4OfficialJul 9, 2025Details
NVIDIA Nemotron 3 UltraOpenNVIDIA87.0%Nemotron-3-Ultra-550B-A55B (BF16)VerifiedJun 4, 2026Details
MiMo-V2.5-ProOpenXiaomi86.6%MiMo-V2.5-ProVerifiedApr 22, 2026Details
GPT-5OpenAI86.2%GPT-5OfficialAug 7, 2025Details
Claude Opus 4.5Anthropic86.0%Claude Opus 4.5OfficialNov 24, 2025Details
GLM-4.7OpenZ.ai85.7%GLM-4.7VerifiedDec 22, 2025Details
GLM-5.1OpenZ.ai85.5%GLM-5.1OfficialApr 7, 2026Details
Gemini 2.5 ProGoogle DeepMind85.3%Gemini 2.5 Pro (Jun 2025)OfficialMar 25, 2025Details
MiniMax M2.5OpenMiniMax85.2%MiniMax-M2.5VerifiedFeb 12, 2026Details
MAI-Code-1-FlashMicrosoft AI84.6%MAI-Code-1-FlashUnverifiedJun 2, 2026Details
Kimi K2 ThinkingOpenMoonshot AI84.5%Kimi K2 ThinkingVerifiedNov 6, 2025Details
MAI-Thinking-1Microsoft AI84.2%MAI-Thinking-1VerifiedJun 2, 2026Details
OpenAI o3-proOpenAI84%OpenAI o3 proUnverifiedJun 10, 2025Details
DeepSeek V3.2OpenDeepSeek83.4%DeepSeek-V3.2OfficialDec 1, 2025Details
Gemini 3 FlashGoogle DeepMind83.2%Gemini 3 FlashOfficialDec 17, 2025Details
Nex-N2-miniOpenNex AGI82.6%Nex-N2-miniVerifiedJun 2, 2026Details
Claude Sonnet 4.5Anthropic82.3%Claude Sonnet 4.5OfficialSep 29, 2025Details
o3OpenAI81.8%o3OfficialApr 16, 2025Details
Amazon Nova 2 ProAmazon81.4%Nova 2 ProVerifiedDec 2, 2025Details
Qwen3 235B A22B Instruct 2507OpenAlibaba / Qwen80.1%Qwen3-235B-A22B (Jul 2025)OfficialJul 21, 2025Details
DeepSeek V3.1OpenDeepSeek80.1%DeepSeek-V3.1-ThinkingVerifiedAug 21, 2025Details
o4-miniOpenAI79.6%o4-miniOfficialApr 16, 2025Details
Amazon Nova 2 LiteAmazon79.6%Nova 2 LiteVerifiedDec 2, 2025Details
NVIDIA Nemotron 3 Super 120B-A12BOpenNVIDIA79.23%N-3-SuperVerifiedMar 10, 2026Details
K-EXAONEOpenLG AI Research79.1%K-EXAONE (Reasoning)UnverifiedJan 12, 2026Details
Claude 3.7 SonnetAnthropic78.5%Claude 3.7 SonnetOfficialFeb 24, 2025Details
Claude Sonnet 4Anthropic78.3%Claude Sonnet 4OfficialMay 22, 2025Details
Claude Opus 4.1Anthropic77.3%Claude Opus 4.1OfficialAug 5, 2025Details
Claude Opus 4Anthropic76.3%Claude Opus 4OfficialMay 22, 2025Details
GPT-OSS-120BOpenOpenAI75.8%gpt-oss-120bOfficialAug 5, 2025Details
GPT-5 miniOpenAI75.0%GPT-5 miniOfficialAug 7, 2025Details
Qwen3 MaxAlibaba / Qwen72.6%Qwen3-MaxOfficialSep 5, 2025Details
Claude Haiku 4.5Anthropic71.2%Claude Haiku 4.5OfficialOct 15, 2025Details
Hunyuan-T1Tencent69.3%Hunyuan-T1VerifiedMar 21, 2025Details
DeepSeek R1OpenDeepSeek69.2%DeepSeek-R1OfficialJan 20, 2025Details
Llama 4 MaverickOpenMeta67.0%Llama 4 MaverickOfficialApr 5, 2025Details
GPT-4.1OpenAI66.9%GPT-4.1OfficialApr 14, 2025Details
Gemini 2.0 FlashGoogle DeepMind64.1%Gemini 2.0 FlashOfficialDec 11, 2024Details
OLMo 3.1-Think 32BOpenAllen Institute for AI57.5%Olmo 3.1 32B ThinkUnverifiedDec 12, 2025Details
Gemini 1.5 ProGoogle DeepMind57.2%Gemini 1.5 ProOfficialFeb 15, 2024Details
DeepSeek V3OpenDeepSeek56.5%DeepSeek-V3OfficialDec 26, 2024Details
Claude 3.5 SonnetAnthropic54.0%Claude 3.5 SonnetOfficialJun 20, 2024Details
Reka Flash 3OpenReka AI52.9%Reka Flash 3UnverifiedMar 10, 2025Details
Llama 4 ScoutOpenMeta51.8%Llama 4 ScoutOfficialApr 5, 2025Details
Llama 3.1 405BOpenMeta50.9%Llama 3.1-405BOfficialJul 23, 2024Details
GPT-4oOpenAI49.2%GPT-4oOfficialMay 13, 2024Details
IBM Granite 4.1OpenIBM41.96%Granite-4.1-8B-InstructUnverifiedApr 29, 2026Details
Jamba 1.7 LargeOpenAI21 Labs39%Jamba 1.7 LargeUnverifiedJul 3, 2025Details
Mistral LargeMistral AI38.8%Mistral LargeOfficialFeb 26, 2024Details

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