# Claude Opus 5.5 can now rebuild real programs from scratch — 18.5% on ProgramBench, up from 0% in May — retained record

> Retained reddit record from reddit:artificial.

Canonical: https://highsignal.app/data/records/a04a23533859826b

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:artificial)  published 2026-10-28 01:46 ingested 2026-10-29 02:36

[original source ↗](https://www.reddit.com/r/artificial/comments/1xhbe9u/claude_opus_55_can_now_rebuild_real_programs_from/)

## retained text



```
According to Jack Clark's Import AI (Oct 26), Claude Opus 5.5 (xhigh) now resolves 18.5% of ProgramBench's 200 from-scratch rebuild tasks — rewriting programs like SQLite, FFmpeg and ripgrep from a compiled binary and its docs. That is up from 0% when every model failed at the benchmark's May launch. Clark expects the benchmark to be saturated by the end of 2027 (his prediction, not a measured result).

What I am curious about: is 'rebuild from binary + docs' mainly a test of reverse-engineering and long-horizon persistence, or does it genuinely measure software engineering ability? And if this benchmark saturates within a year, what should replace it as a meaningful test of coding?
```



document fetched 2026-10-29 00:20 · canonical https://reddit.com/r/artificial/comments/1xhbe9u/claude_opus_55_can_now_rebuild_real_programs_from
