Atlas

Benchmarks

← All benchmarks

OpenAI MRCR v2 8-needle - 4K-8K

General QA

Eight-needle retrieval averaged within the 4K-8K context band. Same OpenAI task family as longer-context bins, distinct from Google's cumulative metric; exact item count unreported.

Top models (higher is better)

ModelScore
GPT-5.598.1
Loading Atlas data…