LibEvoBench - Stable APIs: leaderboard
LibEvoBench (2026) tests version-specific API knowledge of code models across 29 versions of PyTorch, NumPy and SciPy with API calling, API identification and signature recall tasks, averaging four task metrics on a 0-100 scale. This board is the average score on stable APIs, whose signature is identical in every covered version; higher is better.
Metric: Average task performance (0-100) on stable APIs (identical signature in every covered version): mean of four task metrics (API calling with version constraint, exact match; API calling parameter recall given name and version; API identification from a redacted docstring, exact match; signature recall F1) averaged over PyTorch, NumPy and SciPy versions, temperature 0 where supported and no thinking budget; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 92.3 | #49 |
| 2 | GPT-5.4 | 92 | #123 |
| 3 | Claude Sonnet 4.6 | 89.3 | #161 |
| 4 | Claude Sonnet 4 | 87.8 | #400 |
| 5 | Gemini 3 Flash (Minimal) | 87.8 | #164 (Gemini 3 Flash) |
| 6 | GPT-4.1 | 84.3 | #495 |
| 7 | GPT-5.1 | 83.8 | #229 |
| 8 | GPT-5 | 83 | #167 |
| 9 | Gemini 2.5 Flash (Non-reasoning) | 81.6 | #453 (Gemini 2.5 Flash) |
| 10 | Qwen 3.5 397B A17B (Non-reasoning) | 80.2 | #246 (Qwen 3.5 397B A17B) |
| 11 | Gemini 2.0 Flash | 79.7 | #704 |
| 12 | Qwen 3.5 122B A10B (Non-reasoning) | 70.5 | #331 (Qwen 3.5 122B A10B) |
| 13 | Qwen 3.5 35B A3B (Non-reasoning) | 63.3 | #482 (Qwen 3.5 35B A3B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=libevobench-stable-apis · How It Works · Data refreshed daily, snapshot 2026-11-13.