Benchmark
MRCR
MRCR (Multi-Round Coreference Resolution) is a synthetic long-context reasoning task where models must navigate long conversations to reproduce specific…
- Modality
- text
- Categories
- long_context, reasoning, general
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 7
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Gemini 1.5 Flash | 0.719 | 2024-05-01 | evidence | |
| Gemini 1.5 Flash 8B | 0.547 | 2024-03-15 | evidence | |
| Gemini 1.5 Pro | 0.826 | 2024-05-01 | evidence | |
| Gemini 2.0 Flash | 0.692 | 2024-12-01 | evidence | |
| Gemini 2.5 Flash | 0.32 | 2025-05-20 | evidence | |
| Gemini 2.5 Pro | 0.93 | 2025-05-20 | evidence | |
| MiMo-V2-Flash | Xiaomi | 0.457 | 2025-12-16 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.