What it measures
Challenges agents to write core Python libraries (like simpy, minitorch, pydantic) from scratch starting from commit 0 and API docs.
Reading the measurement
Look beyond a plausible code sample to the executable artifact. Tests, repository context, language coverage and time limits shape the result. Separate first-attempt correctness from success after debugging, retries or agent assistance.
Published evaluation
Mean unit-test pass rate at the named agent stage. Dollar-cost subscripts are retained in source-cell notes but are not part of the score.
Where this fits in an evaluation suite
The agent wrapper is part of the system. Pin tools, dependencies, timeouts, tests, and the repository revision. Compare contest coding separately from repository maintenance.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
Mean unit-test pass rate at the named agent stage. Dollar-cost subscripts are retained in source-cell notes but are not part of the score.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.
PRIMARY EVIDENCE & FURTHER READING
- Commit0: Library Generation from Scratch — published result table ↗
All 7 authors
Zhao, Wenting, Jiang, Nan, Lee, Celine, Chiu, Justin T, Cardie, Claire, Gallé, Matthias, Rush, Alexander M
Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.