reasoning benchmark
FRAMES
Factuality, Retrieval, And reasoning MEasurement Set - a unified evaluation dataset of 824 challenging multi-hop questions for testing retrieval-augmented generation systems across factuality, retrieval accuracy, and reasoning capabilities, requiring integration of 2-15 Wikipedia articles per question
Updated Aug 7, 2026
Published models2
Registry coverage2
MetricScore
EvidenceB
FRAMES leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | MA | 87.0% | 100.0% | 2 | C | |
| 2 | DE | 73.3% | 0.0% | 2 | C |
Score distribution
Top published rows on the benchmark's original scale.
FRAMES highlights
The top published results on this benchmark's own scale.
What is FRAMES?
Definition and scoring fields from the benchmark registry.
Factuality, Retrieval, And reasoning MEasurement Set - a unified evaluation dataset of 824 challenging multi-hop questions for testing retrieval-augmented generation systems across factuality, retrieval accuracy, and reasoning capabilities, requiring integration of 2-15 Wikipedia articles per question
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- FRAMES
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- frames|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about FRAMES.
Which model scores highest on FRAMES?
Kimi K2-Thinking-0905 is currently ranked first with 87.0%.
What does FRAMES measure?
Factuality, Retrieval, And reasoning MEasurement Set - a unified evaluation dataset of 824 challenging multi-hop questions for testing retrieval-augmented generation systems across factuality, retrieval accuracy, and reasoning capabilities, requiring integration of 2-15 Wikipedia articles per question
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
2 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.