evals.report
BenchmarksLabsCompareRun guidesIn the wild
BenchmarksReasoning

FACTS Grounding

A Google DeepMind benchmark that measures how factually grounded an LLM's long-form responses are to a provided source document, scoring the share of responses that are eligible and fully supported by the context with no hallucinations.

ReasoningGrounding accuracyHigher is better
ScoresAboutRun this benchmark

What is FACTS Grounding?

A Google DeepMind benchmark that measures how factually grounded an LLM's long-form responses are to a provided source document, scoring the share of responses that are eligible and fully supported by the context with no hallucinations. evals.report tracks reported FACTS Grounding scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.

Top reported FACTS Grounding score: Gemini 2.0 Flash — 83.6% (Grounding accuracy).

ModelLabScore↓Source modelStatusDate
Gemini 2.0 FlashGoogle DeepMind83.6%—VerifiedDec 11, 2024Details
Gemini 1.5 ProGoogle DeepMind80.0%—VerifiedFeb 15, 2024Details
Claude 3.5 SonnetAnthropic79.4%—VerifiedJun 20, 2024Details
GPT-4oOpenAI78.8%—VerifiedMay 13, 2024Details
Gemini 2.5 ProGoogle DeepMind74.2%—VerifiedMar 25, 2025Details
Gemini 2.5 FlashGoogle DeepMind69.9%—VerifiedApr 17, 2025Details
GPT-5OpenAI69.6%—VerifiedAug 7, 2025Details
Gemini 3 ProGoogle DeepMind69.0%—VerifiedNov 18, 2025Details
Claude Opus 4.5Anthropic62.1%—VerifiedNov 24, 2025Details
Claude Sonnet 4.5Anthropic61.8%—VerifiedSep 29, 2025Details
GPT-5 miniOpenAI58.3%—VerifiedAug 7, 2025Details
Claude Sonnet 4Anthropic56.1%—VerifiedMay 22, 2025Details
Claude Opus 4.1Anthropic54.8%—VerifiedAug 5, 2025Details
Grok 4xAI54.7%—VerifiedJul 9, 2025Details
GPT-5.1OpenAI50.0%—VerifiedNov 12, 2025Details
GPT-4.1OpenAI45.6%—VerifiedApr 14, 2025Details
o3OpenAI36.2%—VerifiedApr 16, 2025Details
o4-miniOpenAI29.3%—VerifiedApr 16, 2025Details

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