HELM: leaderboard
HELM: Evaluates broad language-model knowledge, reasoning, commonsense, instruction following, or exam-style accuracy.
Metric: Mean win rate (self-reported). Source: benchmarklist.com. Status: saturated. 66 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 2 70B | 94.35 |
| 2 | LLaMA-65B | 90.83 |
| 3 | text-davinci-002 | 90.5 |
| 4 | Mistral-7B-v0.1 | 88.4 |
| 5 | text-davinci-003 | 87.16 |
| 6 | Llama 2 13B | 82.3 |
| 7 | GPT-3.5 Turbo (0613) | 78.3 |
| 8 | LLaMA-30B | 78.13 |
| 9 | GPT-3.5 Turbo (0301) | 76.03 |
| 10 | falcon-40B | 72.94 |
| 11 | mpt-30B | 71.45 |
| 12 | Llama 2 7B | 60.73 |
| 13 | LLaMA-13B | 59.47 |
| 14 | davinci | 53.77 |
| 15 | LLaMA-7B | 53.27 |
Interactive version: theaggregate.ai/benchmark?slug=helm · How It Works · Data refreshed daily, snapshot 2026-09-05.