Atlas

Benchmarks

← All benchmarks

OpenAI MRCR v2 8-needle - 256K-512K

General QA

OpenAI multi-round coreference retrieval with eight needles, averaged within the 256K-512K context band. Distinct from Google's cumulative MRCR and its 1M point evaluation; exact item count is unreported.

Top models (higher is better)

ModelScore
GPT-6 Astra100.0
GPT-5.6 Sol91.5
GPT-5.6 Terra89.6
GPT-5.581.5
GPT-5.6 Luna41.3
Loading Atlas data…