57. Ran the frontier models on my long context benchmark - LongCodeEdit, now extended to 512K context from 128K. by @nrehiew_ (wh) · backlist 2026-05-08 · rubric 89.0