all the models — AI benchmark observatory
← Benchmarks

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.

id mrcr · max 1 · 7 models reported

#ModelScore
1Gemini 2.5 Pro
Google
0.93
MRCR Leaderboard · all the models