Claude Opus 4.8 Hits Record 16.5% on ProgramBench

Claude Opus 4.8 achieved a record 16.5% "almost resolved" score on ProgramBench, narrowly missing full marks on three tasks. Compared to GLM 5.2 and earlier models, Opus 4.8 also demonstrated superior efficiency by achieving these top scores using fewer steps.

2026-07-25 ~ 2026-07-25 · 3 related posts

• Claude Opus 4.8 hits 16.5% on ProgramBench’s almost-resolved metric — jyangballin · 2026-07-25
• Claude Opus 4.8 missed full marks by tiny details on three ProgramBench tasks — jyangballin · 2026-07-25
• Claude Opus 4.8 reaches its best ProgramBench score in fewer steps — jyangballin · 2026-07-25