SEAL - Math: leaderboard
GSM1k dataset of 1,000 human-written grade-school math word problems evaluating multi-step arithmetic reasoning, created to mirror GSM8k difficulty while avoiding contamination.
Metric: Score. Source: scale.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20240620) | 96.6 | #299 |
| 2 | GPT-4o (2024-08-06) | 95.68 | #336 |
| 3 | Llama 3.1 405B Instruct | 95.6 | #457 |
| 4 | Claude 3 Opus | 95.19 | #571 |
| 5 | GPT-4 Turbo (Preview) | 95.1 | #397 |
| 6 | GPT-4o (2024-05-13) | 94.85 | #325 |
| 7 | Mistral Large 2 (Jul) | 93.94 | #484 |
| 8 | Claude 3 Sonnet | 93.28 | #775 |
| 9 | Gemini 1.5 Pro (0514) | 92.28 | #512 |
| 10 | Llama 3 70B Instruct | 90.12 | #634 |
| 11 | Gemini 1.5 Flash | 90.12 | #618 |
| 12 | Mistral Large | 87.47 | #710 |
| 13 | Gemini 1.0 Pro | 79.83 | #933 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-math · How It Works · Data refreshed daily, snapshot 2026-10-20.