Python subset of CodeXGLUE Code-to-Text: generate natural-language docstrings/comments from Python function bodies. Derived from CodeSearchNet. Evaluated with smoothed BLEU-4.
Bleu is the reported evaluation metric for CodeXGLUE Code-to-Text (Python). Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.
Higher is better
| Rank | Model | Trust | Score | Year | Links | Fix |
|---|---|---|---|---|---|---|
| 01 | CodeT5-base | paper | 20.01 | 2026 | Source ↗ | Looks wrong? |
| 02 | PLBART | paper | 19.3 | 2026 | Source ↗ | Looks wrong? |
| 03 | CodeBERT | paper | 19.06 | 2026 | Source ↗ | Looks wrong? |