Benchmark
MRCR Benchmark
What is MRCR?
Needle position and generation seed change the result. Benchmark Radar tracks 7 reported scores for MRCR, cited by 8 source documents.
- Released
- 2025-04-14
- Evaluates
- long context
- Openness
- unknown
- Importer
- Model reports
- Reported scores
- 7
Source provenance
- Original evidence https://huggingface.co/datasets/openai/mrcr
MRCR links
MRCR results and reported scores
Model reports
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| DeepSeek-V4-Flash | DeepSeek | 78.7 | 2026-06-25 | evidence |
| DeepSeek-V4-Pro | DeepSeek | 44.7 | 2026-04-22 | evidence |
| DeepSeek-V4-Pro | DeepSeek | 83.3 | 2026-04-22 | evidence |
| DeepSeek-V4-Pro | DeepSeek | 83.5 | 2026-04-22 | evidence |
| Gemini 3.1 Pro | 84.9 | 2026-02-19 | evidence | |
| Gemini 3.1 Pro | 26.3 | 2026-02-19 | evidence | |
| Gemma 4 (31B) | 66.4 | 2026-03-11 | 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.
Which sources cite MRCR?
- DeepSeek-V4 DeepSeek, 2026-06-25
- DeepSeek-V4-Pro DeepSeek, 2026-04-22
- Gemma 4 (31B) Google, 2026-03-11
- Gemini 3.1 Pro Google, 2026-02-19
- GPT-5 OpenAI, 2025-08-07
- Gemini 2.5 Pro Google, 2025-06-17
- Claude Opus 4 and Sonnet 4 Anthropic, 2025-05-22
- GPT-4.1 OpenAI, 2025-04-14