Natural2Code: leaderboard
Google's internal held-out Python code-generation set in HumanEval format, author-written to avoid web leakage, reported 0-shot in the Gemini 1.0 report (2023): Gemini Ultra 74.9%, GPT-4 73.9%.
Metric: Score (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 52.8 | #438 |
| 2 | GPT-4 Turbo (Preview) | 51.5 | #397 |
| 3 | Claude 3 Opus | 48.3 | #571 |
| 4 | Gemini 1.5 Pro | 42.3 | #445 |
| 5 | GPT-3.5 Turbo | 40.7 | #865 |
| 6 | Claude 3 Sonnet | 38.9 | #775 |
| 7 | Claude 3 Haiku | 36.2 | #803 |
| 8 | Claude 2.1 | 34.4 | #839 |
| 9 | DeepSeek V3 Chat | 33.2 | #447 |
| 10 | starcoder | 13.2 | #1577 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=natural2code · How It Works · Data refreshed daily, snapshot 2026-10-20.