Code Generation Eval

Sample data

Code generation

Definition & methodology

Methodology
Sample methodology: models generate solutions to programming tasks, graded by unit tests.
Dataset
Sample dataset of function-level programming problems with hidden test suites.
Metric
pass@1 (%) (higher is better)
Version
1.2
Last updated
2026-08-01
Limitations
Sample limitation note: limited language coverage in this demo dataset.

Model results

Results by model for Code Generation Eval
ModelScoreSample sizeEvidenceEvaluated
GPT Frontier80 %300L2Benchmark Result2026-08-01
Claude Sonnet78 %300L2Benchmark Result2026-08-01
Gemini Pro75 %300L2Benchmark Result2026-08-01