EffiBench - NMU: leaderboard
EffiBench normalized memory usage (NMU) — memory efficiency of model-generated programs relative to reference solutions; lower is better.
Metric: Normalized Memory Usage. Source: huggingface.co. 42 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | starcoder2-15B | 1.71 | #1495 |
| 2 | Yi Chat 34B | 1.89 | #976 |
| 3 | Yi 34B (Base) | 1.89 | #972 |
| 4 | Phind-CodeLlama-34B-v2 | 1.89 | #1589 |
| 5 | deepseek-coder-6.7B-instruct | 1.9 | #1335 |
| 6 | Claude 3 Haiku | 1.91 | #803 |
| 7 | Claude 3 Sonnet | 1.91 | #775 |
| 8 | starcoder | 1.91 | #1577 |
| 9 | WizardCoder-15B-V1.0 | 1.91 | #1588 |
| 10 | deepseek-coder-1.3B-instruct | 1.91 | #1585 |
| 11 | Magicoder-S-DS-6.7B | 1.91 | #1256 |
| 12 | deepseek-coder-6.7B-base | 1.91 | #1484 |
| 13 | GPT-3.5 Turbo (0613) | 1.92 | #1118 |
| 14 | GPT-4 | 1.92 | #438 |
| 15 | CodeLlama-34B-hf | 1.92 | #1515 |
Interactive version: theaggregate.ai/benchmark?slug=effibench-nmu · How It Works · Data refreshed daily, snapshot 2026-10-20.