HELM Lite - GSM8K: leaderboard
Stanford CRFM HELM Lite (release v1.13.0) run of GSM8K grade-school math word problems: 1,000 test problems, 5-shot, scored by exact match of the final answer. Percent correct, 0-100, higher is better.
Metric: Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 90 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 95.6 | #302 |
| 2 | Claude 3.5 Sonnet (20240620) | 94.9 | #308 |
| 3 | Llama 3.1 405B Instruct | 94.9 | #470 |
| 4 | Gemini 2.0 Flash (Preview) | 94.6 | #376 |
| 5 | Llama 3.3 70B Instruct | 94.2 | #541 |
| 6 | DeepSeek V3 | 94 | #329 |
| 7 | Llama 3.1 70B Instruct | 93.8 | #567 |
| 8 | Llama 3.2 90B Vision Instruct | 93.6 | #621 |
| 9 | GPT-4 (0613) | 93.2 | #493 |
| 10 | Claude 3 Opus (20240229) | 92.4 | #428 |
| 11 | Qwen 2 72B Instruct | 92 | #579 |
| 12 | Mistral Large 2 (Jul) | 91.2 | #490 |
| 13 | GPT-4o (2024-08-06) | 90.9 | #343 |
| 14 | Claude 3 Sonnet (20240229) | 90.7 | #694 |
| 15 | GPT-4o (2024-05-13) | 90.5 | #337 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-lite-gsm8k · How It Works · Data refreshed daily, snapshot 2026-11-05.