reasoning benchmark
MRCR
MRCR (Multi-Round Coreference Resolution) is a synthetic long-context reasoning task where models must navigate long conversations to reproduce specific model outputs. It tests the ability to distinguish between similar requests and reason about ordering while maintaining attention across extended contexts.
Updated Aug 7, 2026
Published models7
Registry coverage7
MetricScore
EvidenceB
MRCR leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | GO | 93.0% | 100.0% | 7 | C | |
| 2 | GO | 82.6% | 83.3% | 7 | C | |
| 3 | GO | 71.9% | 66.7% | 7 | C | |
| 4 | GO | 69.2% | 50.0% | 7 | C | |
| 5 | GO | 54.7% | 33.3% | 7 | C | |
| 6 | XI | 45.7% | 16.7% | 7 | C | |
| 7 | GO | 32.0% | 0.0% | 7 | C |
Score distribution
Top published rows on the benchmark's original scale.
MRCR highlights
The top published results on this benchmark's own scale.
What is MRCR?
Definition and scoring fields from the benchmark registry.
MRCR (Multi-Round Coreference Resolution) is a synthetic long-context reasoning task where models must navigate long conversations to reproduce specific model outputs. It tests the ability to distinguish between similar requests and reason about ordering while maintaining attention across extended contexts.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- MRCR
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- mrcr|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about MRCR.
Which model scores highest on MRCR?
Gemini 2.5 Pro is currently ranked first with 93.0%.
What does MRCR measure?
MRCR (Multi-Round Coreference Resolution) is a synthetic long-context reasoning task where models must navigate long conversations to reproduce specific model outputs. It tests the ability to distinguish between similar requests and reason about ordering while maintaining attention across extended contexts.
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.