Atlas

Benchmarks

← All benchmarks

OpenAI MRCR v2 8-needle - 512K-1M

General QA

Eight-needle OpenAI MRCR v2 retrieval averaged over the 512K-1M context band. Shares a factor group with the preceding band and is not Google's 1M pointwise metric.

Top models (higher is better)

ModelScore
GPT-6 Astra96.3
GPT-5.574.0
GPT-5.6 Sol73.8
GPT-5.6 Terra72.5
GPT-5.6 Luna41.3
Loading Atlas data…