evals.report
BenchmarksLabsCompareRun guides
BenchmarksAgents

GDPval

GDPval evaluates AI models agentically (shell + web access via a sandbox harness) on real-world economically valuable knowledge-work deliverables — documents, spreadsheets, slides, diagrams — spanning 44 occupations across 9 major U.S. GDP industries, scored by blind pairwise quality comparison; the Artificial Analysis GDPval-AA variant reports results as an Elo rating.

AgentsEloHigher is better
ScoresAboutRun this benchmark
ModelLabScore↓Source modelStatusDate
Claude Opus 4.8Anthropic1890—OfficialMay 28, 2026Details
GPT-5.5OpenAI1769—OfficialApr 23, 2026Details
Claude Opus 4.7Anthropic1753—OfficialApr 16, 2026Details
Claude Sonnet 4.6Anthropic1676—OfficialFeb 17, 2026Details
GPT-5.4OpenAI1674—OfficialMar 5, 2026Details
Gemini 3.5 FlashGoogle DeepMind1659—OfficialMay 19, 2026Details
Claude Opus 4.6Anthropic1619—OfficialFeb 5, 2026Details
MiMo-V2.5-ProXiaomi1571—OfficialApr 22, 2026Details
DeepSeek V4 ProDeepSeek1558—OfficialApr 24, 2026Details
MiMo-V2.5Xiaomi1551—OfficialApr 22, 2026Details
GLM-5.1Z.ai1535—OfficialApr 7, 2026Details
MiniMax M2.7MiniMax1505—OfficialMar 18, 2026Details
Qwen 3.6 Max PreviewAlibaba / Qwen1504—OfficialApr 20, 2026Details
Grok 4.3xAI1495—OfficialApr 17, 2026Details
GPT-5.3-CodexOpenAI1482—OfficialFeb 5, 2026Details
Kimi K2.6Moonshot AI1481—OfficialApr 20, 2026Details
GPT-5.2OpenAI1467—OfficialDec 11, 2025Details
Claude Opus 4.5Anthropic1452—OfficialNov 24, 2025Details
Muse SparkMeta1417—OfficialApr 8, 2026Details
DeepSeek V4 FlashDeepSeek1414—OfficialApr 24, 2026Details
GLM-5Z.ai1395—OfficialFeb 11, 2026Details
Qwen 3.6 PlusAlibaba / Qwen1354—OfficialApr 2, 2026Details
Gemini 3 Deep ThinkGoogle DeepMind1324—OfficialDec 4, 2025Details
Claude Sonnet 4.5Anthropic1317—OfficialSep 29, 2025Details
Gemini 3.1 Pro PreviewGoogle DeepMind1314—OfficialFeb 19, 2026Details
Step 3.7 FlashStepFun1298—OfficialMay 29, 2026Details
GPT-5OpenAI1294—OfficialAug 7, 2025Details
GPT-5.2-CodexOpenAI1288—OfficialDec 18, 2025Details
Kimi K2.5Moonshot AI1285—OfficialJan 27, 2026Details
GPT-5.1OpenAI1227—OfficialNov 12, 2025Details
Qwen3.5-397B-A17BAlibaba / Qwen1220—OfficialFeb 16, 2026Details
Gemini 3 FlashGoogle DeepMind1204—OfficialDec 17, 2025Details
DeepSeek V3.2DeepSeek1197—OfficialDec 1, 2025Details
GLM-4.7Z.ai1185—OfficialDec 22, 2025Details
GPT-5 miniOpenAI1184—OfficialAug 7, 2025Details
Gemini 3 ProGoogle DeepMind1184—OfficialNov 18, 2025Details
MiniMax M2.5MiniMax1176—OfficialFeb 12, 2026Details
Claude Haiku 4.5Anthropic1171—OfficialOct 15, 2025Details
Mistral Medium 3.5Mistral AI1168—OfficialApr 28, 2026Details
Claude Sonnet 4Anthropic1133—OfficialMay 22, 2025Details
MiniMax M2.1MiniMax1091—OfficialDec 23, 2025Details
DeepSeek V3.1DeepSeek1080—OfficialAug 21, 2025Details
Gemini 2.5 FlashGoogle DeepMind1071—OfficialApr 17, 2025Details
Claude 3.7 SonnetAnthropic1048—OfficialFeb 24, 2025Details
Grok 4.1 fast reasoningxAI1046—OfficialNov 19, 2025Details
Qwen3 MaxAlibaba / Qwen1038—OfficialSep 5, 2025Details
GLM-4.6Z.ai1029—OfficialSep 30, 2025Details
o4-miniOpenAI1008—OfficialApr 16, 2025Details
Kimi K2 ThinkingMoonshot AI992—OfficialNov 6, 2025Details
Grok 4xAI989—OfficialJul 9, 2025Details
GPT-OSS-120BOpenAI947—OfficialAug 5, 2025Details
Gemini 2.5 ProGoogle DeepMind919—OfficialMar 25, 2025Details
Mistral LargeMistral AI864—OfficialFeb 26, 2024Details
K-EXAONELG AI Research825—OfficialJan 12, 2026Details
Qwen3 235B A22B Instruct 2507Alibaba / Qwen778—OfficialJul 21, 2025Details
GPT-4.1OpenAI776—OfficialApr 14, 2025Details
o3OpenAI753—OfficialApr 16, 2025Details
Gemini 2.0 FlashGoogle DeepMind566—OfficialDec 11, 2024Details
Qwen 3 Coder 480BAlibaba / Qwen506—OfficialJul 22, 2025Details
Solar Pro 2Upstage449—OfficialJul 10, 2025Details
Llama 4 MaverickMeta435—OfficialApr 5, 2025Details
DeepSeek V3 0324DeepSeek407—OfficialMar 24, 2025Details
GPT-4oOpenAI378—OfficialMay 13, 2024Details
Jamba 1.7 LargeAI21 Labs282—OfficialJul 3, 2025Details
Llama 4 ScoutMeta270—OfficialApr 5, 2025Details
Llama 3.1 405BMeta255—OfficialJul 23, 2024Details
DeepSeek R1DeepSeek248—OfficialJan 20, 2025Details

Each row reports the model’s Elo on GDPval. Click a row for the full run context.