Claude Opus 5 Tops ProgramBench with Massive Leap in Zero-Shot Program Building
jyangballin · x · 2026-08-14
A recent tweet discussed the latest progress on ProgramBench, a highly challenging benchmark requiring AI agents to rebuild whole programs (like sqlite, ffmpeg, php) from scratch.
When initially launched, models could only solve 2 out of 200 tasks. However, the newly evaluated Claude Opus 5 (xhigh) made an enormous jump, fully resolving 9 task instances (a 4.5% solve rate). It is now the clear #1 on the leaderboard, demonstrating a significant capability leap in complex coding tasks.
Related event: Claude Opus 5 Tops ProgramBench Coding Benchmark(7 posts)→
More from coding & agent
- Claude Code and Codex now autocomplete the user's next action — tokenbender · 2026-10-16
- Hermes Agent ships official Reverse Engineer Anything plugin with one-command setup — Teknium · 2026-10-16
- MLExpert: 35B MoE Agent Trained to Autonomously Handle Machine Learning Engineering — Shangheng Du · 2026-10-16
- Cowork-Gym: Alibaba Synthesizes 64.9K Agent Work Environments, Boosting Qwen by 8.7 Points — Anni Zou · 2026-10-16
- MMSearch-OPD: Joint RL and On-Policy Distillation for Long-Horizon Multimodal Search Agents — PekingUniversity · 2026-10-16
- A gallery of 214 projects built on one model shows the 'post your build' ad flywheel — gaganghotra_ · 2026-10-16