reasoning benchmark
Natural Questions
Natural Questions is a question answering dataset featuring real anonymized queries issued to Google search engine. It contains 307,373 training examples where annotators provide long answers (passages) and short answers (entities) from Wikipedia pages, or mark them as unanswerable.
Updated Aug 7, 2026
Published models7
Registry coverage7
MetricScore
EvidenceB
Natural Questions leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | GO | 34.5% | 100.0% | 7 | C | |
| 2 | MA | 31.2% | 83.3% | 7 | C | |
| 3 | GO | 29.2% | 66.7% | 7 | C | |
| 4 | GO | 20.9% | 50.0% | 7 | C | |
| 5 | GO | 20.9% | 33.3% | 7 | C | |
| 6 | GO | 15.5% | 16.7% | 7 | C | |
| 7 | GO | 15.5% | 0.0% | 7 | C |
Score distribution
Top published rows on the benchmark's original scale.
Natural Questions highlights
The top published results on this benchmark's own scale.
What is Natural Questions?
Definition and scoring fields from the benchmark registry.
Natural Questions is a question answering dataset featuring real anonymized queries issued to Google search engine. It contains 307,373 training examples where annotators provide long answers (passages) and short answers (entities) from Wikipedia pages, or mark them as unanswerable.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- Natural Questions
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- natural-questions|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about Natural Questions.
Which model scores highest on Natural Questions?
Gemma 2 27B is currently ranked first with 34.5%.
What does Natural Questions measure?
Natural Questions is a question answering dataset featuring real anonymized queries issued to Google search engine. It contains 307,373 training examples where annotators provide long answers (passages) and short answers (entities) from Wikipedia pages, or mark them as unanswerable.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
7 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.