Big Code Memorization: leaderboard
Code-generation evaluation from the Memorization or Generation of Big Code Models study, comparing HumanEval solutions with transformed HumanEval-ET variants that stress generalization.
Metric: HumanEval pass@1 (%). Source: huggingface.co. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | deepseek-coder-6.7B-instruct | 71.34 |
| 2 | GPT-3.5 Turbo (0613) | 65.85 |
| 3 | starcoder2-15B | 46.95 |
| 4 | CodeQwen1.5-7B | 43.9 |
| 5 | starcoder2-7B | 36.59 |
Interactive version: theaggregate.ai/benchmark?slug=big-code-memorization · How It Works · Data refreshed daily, snapshot 2026-09-05.