ProgramBench S
AI benchmark · Software Engineering
- Publisher
- Anthropic System Card
- Domains
- Software Engineering
- Tier
- S why
- Snapshot
Full listing name: ProgramBench (Anthropic Harness)
What is ProgramBench?
Anthropic system-card ProgramBench evaluation on the 166-task golden subset, reported as hidden behavioral test pass rate across 1-5 episodes.
Tier S
The standard: frontier labs report it, or a trusted evaluator scores hundreds of models on it.
How much stands behind a tier: usage, who backs it, a confirmed line from its own site or paper, and a clear read on X. High with three of the four, medium with two, low otherwise.
- 37 models scored
- reported by Anthropic, Meta
- in curated RL directories (2 pts)
On X · mixed, moderate buzz, from 13 posts about ProgramBench in the last month
ProgramBench is used for leaderboards and research papers, but critics cite easy reward hacks, a strict 100% threshold, unspecified tests, and confusing score discrepancies.
How we got to tier S
- Adoption: 37 models scored; reported by Anthropic, Meta. Tier S.
- On X: mixed, moderate buzz. “ProgramBench is used for leaderboards and research papers, but critics cite easy reward hacks, a strict 100% threshold, unspecified tests, and confusing score discrepancies.” No move.
- In curated RL directories: 2 points, a floor of C.
- Not yet confirmed by a line from its own site or paper.
A clear view on X moves a tier one step up or down; most entries don't move. How tiers work.
Not on the map yet: no line from its own site or paper places it in a domain.
Next: browse all benchmarks. Entry from the RL Research daily scrape of public sources, 2026-10-11 snapshot.