Claude Opus 5 Tops ProgramBench, Significantly Beating GPT-5.6

jyangballin · x · 2026-08-13

On the ProgramBench, which asks AI to rebuild whole programs (like sqlite, ffmpeg) from scratch, Claude Opus 5 (xhigh) is the new #1 by resolving 9 tasks (4.5%). It significantly outperforms the previous high set by GPT 5.6 Sol, which only resolved 2 tasks.

Related event: Claude Opus 5 Tops ProgramBench but Costs Over $50 Per Task(6 posts)→

Original post →

More from Models

Models channel →