evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksAgents

GAIA: A Benchmark for General AI Assistants

GAIA is a benchmark of 450+ real-world questions requiring multi-step reasoning, web browsing, multi-modality handling, and tool use, designed to be easy for humans (~92%) but hard for AI assistants, scored across three difficulty levels.

AgentsaccuracyHigher is better
ScoresAboutRun this benchmark

What is GAIA: A Benchmark for General AI Assistants?

GAIA is a benchmark of 450+ real-world questions requiring multi-step reasoning, web browsing, multi-modality handling, and tool use, designed to be easy for humans (~92%) but hard for AI assistants, scored across three difficulty levels. evals.report tracks reported GAIA: A Benchmark for General AI Assistants scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported GAIA: A Benchmark for General AI Assistants score: Claude Sonnet 4.5 — 74.55% (accuracy).

ModelLabScore↓Source modelStatusDate
Claude Sonnet 4.5Anthropic74.55%—UnverifiedSep 29, 2025Details
Claude Opus 4.1Anthropic68.48%—UnverifiedAug 5, 2025Details
Claude Opus 4Anthropic64.85%—UnverifiedMay 22, 2025Details
Claude Haiku 4.5Anthropic56.36%—UnverifiedOct 15, 2025Details
Claude Mythos PreviewAnthropic52.3%—UnverifiedApr 7, 2026Details
GPT-5.4 ProOpenAI50.5%—UnverifiedMar 5, 2026Details
GPT-4.1OpenAI50.30%—UnverifiedApr 14, 2025Details
GPT-5.4OpenAI48.2%—UnverifiedMar 5, 2026Details
Claude Opus 4.6Anthropic47.8%—UnverifiedFeb 5, 2026Details
Gemini 3.1 Pro PreviewGoogle DeepMind46.1%—UnverifiedFeb 19, 2026Details
Claude Sonnet 4.6Anthropic45.5%—UnverifiedFeb 17, 2026Details
GPT-5 miniOpenAI44.8%—UnverifiedAug 7, 2025Details
Claude 3.7 SonnetAnthropic43.9%—UnverifiedFeb 24, 2025Details
GPT-5OpenAI42.1%—UnverifiedAug 7, 2025Details
GPT-5.2OpenAI40.3%—UnverifiedDec 11, 2025Details
Gemini 3 ProGoogle DeepMind38.5%—UnverifiedNov 18, 2025Details
Kimi K2.5OpenMoonshot AI38.1%—UnverifiedJan 27, 2026Details
Qwen 3.6 PlusAlibaba / Qwen37.4%—UnverifiedApr 2, 2026Details
o4-miniOpenAI36.8%—UnverifiedApr 16, 2025Details
Gemini 3 FlashGoogle DeepMind35.2%—UnverifiedDec 17, 2025Details
DeepSeek V3.2OpenDeepSeek34.8%—UnverifiedDec 1, 2025Details
GLM-5OpenZ.ai33.8%—UnverifiedFeb 11, 2026Details
Gemini 2.5 ProGoogle DeepMind33.3%—UnverifiedMar 25, 2025Details
o3OpenAI32.73%—UnverifiedApr 16, 2025Details
Gemini 2.0 FlashGoogle DeepMind32.73%—UnverifiedDec 11, 2024Details
DeepSeek R1OpenDeepSeek30.30%—UnverifiedJan 20, 2025Details
DeepSeek V3OpenDeepSeek29.39%—UnverifiedDec 26, 2024Details
Llama 4 MaverickOpenMeta28.6%—UnverifiedApr 5, 2025Details
DeepSeek V3.1OpenDeepSeek11.5%—UnverifiedAug 21, 2025Details

Each row reports the model’s accuracy on GAIA: A Benchmark for General AI Assistants. Click a row for the full run context.