reasoning benchmark
FACTS Grounding
A benchmark evaluating language models' ability to generate factually accurate and well-grounded responses based on long-form input context, comprising 1,719 examples with documents up to 32k tokens requiring detailed responses that are fully grounded in provided documents
Updated Aug 7, 2026
Published models13
Registry coverage13
MetricScore
EvidenceC
FACTS Grounding leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | GO | 87.8% | 100.0% | 13 | C | |
| 2 | GO | 85.3% | 91.7% | 13 | C | |
| 3 | GO | 84.1% | 83.3% | 13 | C | |
| 4 | GO | 83.6% | 75.0% | 13 | C | |
| 5 | GO | 83.6% | 66.7% | 13 | C | |
| 6 | GO | 75.8% | 58.3% | 13 | C | |
| 7 | GO | 74.9% | 50.0% | 13 | C | |
| 8 | GO | 70.5% | 41.7% | 13 | C | |
| 9 | GO | 70.1% | 33.3% | 13 | C | |
| 10 | GO | 61.9% | 25.0% | 13 | C | |
| 11 | ZA | 58.6% | 16.7% | 13 | C | |
| 12 | GO | 40.6% | 8.3% | 13 | C | |
| 13 | GO | 36.4% | 0.0% | 13 | C |
Score distribution
Top published rows on the benchmark's original scale.
FACTS Grounding highlights
The top published results on this benchmark's own scale.
What is FACTS Grounding?
Definition and scoring fields from the benchmark registry.
A benchmark evaluating language models' ability to generate factually accurate and well-grounded responses based on long-form input context, comprising 1,719 examples with documents up to 32k tokens requiring detailed responses that are fully grounded in provided documents
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level C.
- Family
- FACTS Grounding
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- facts-grounding|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about FACTS Grounding.
Which model scores highest on FACTS Grounding?
Gemini 2.5 Pro Preview 06-05 is currently ranked first with 87.8%.
What does FACTS Grounding measure?
A benchmark evaluating language models' ability to generate factually accurate and well-grounded responses based on long-form input context, comprising 1,719 examples with documents up to 32k tokens requiring detailed responses that are fully grounded in provided documents
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
13 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is marked as eligible for the current LLMBoard capability methodology.