ProgramBench hidden-test pass rate after episode 1 (ProgramBench (episode 1))
We show this table for reference; we do not rank on it.
Program-reconstruction hidden-test pass rate after the first of five sequential long-context episodes.
Benchmark score on ProgramBench (episode 1) — September 27, 2026
We compile the ProgramBench (episode 1) rows from provider self-reports. Claude Opus 5 leads the table at 83.0%. We do not use these results to rank models overall.
1 modelCodingCurrentDisplay onlyUpdated September 27, 2026
Benchmark score table (1 model)
ScoreAbout ProgramBench (episode 1)
Year
2026
Tasks
166 golden program-reconstruction tasks
Format
Hidden-test pass rate after episode 1
Difficulty
Long-context clean-room software engineering
Anthropic evaluated 166 golden tasks with no internet or decompilation tools and a fresh context budget of up to one million tokens per episode. This is kept separate from the episode-5 result.
Freshness and provenance
Version
ProgramBench (episode 1) 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does ProgramBench (episode 1) measure?
Program-reconstruction hidden-test pass rate after the first of five sequential long-context episodes.
Which model scores highest on ProgramBench (episode 1)?
Claude Opus 5 by Anthropic currently leads with a score of 83.0% on ProgramBench (episode 1).
How many models are evaluated on ProgramBench (episode 1)?
1 AI models have been evaluated on ProgramBench (episode 1) on BenchLM.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 2,000+ readers.
One email each week. Unsubscribe anytime.