Benchmark Radar
AI BENCHMARK PROFILE

MRCR 128K (2-needle)

General AILong Context & Memory

MRCR (Multi-Round Coreference Resolution) at 128K context length with 2 needles. Models must navigate long conversations to reproduce specific model outputs, testing attention and reasoning across 128K-token contexts with 2 items to retrieve.

Released
Unknown
Readiness
Paper only
Primary field
General AI

Primary resources

Benchmark Radar records only publicly supported details and links back to primary sources for verification.