MRCR v2 8-needle 512K-1M
3 models tested · Updated 2026-07-09 · Verified sources only
GPT-5.6 Sol leads at 73.8%
1
OpenAI · Blog/OpenAI · 2026-07-09
512K-1M range 8-needle MRCR. Luna collapses to 41.3.
73.8%
2
OpenAI · Blog/OpenAI · 2026-07-09
512K-1M 8-needle MRCR. Holds up better than Luna.
72.5%
3
OpenAI · Blog/OpenAI · 2026-07-09
Luna collapses on 1M-context. Same as 256K score.
41.3%