SEAL - Instruction Following: leaderboard
1,054 human-generated prompts across content writing, role-playing, and brainstorming, scored by annotators on a 7-point scale for precise multi-constraint instruction execution.
Metric: Score. Source: scale.com. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 (2024-12-17) | 91.96 | #149 |
| 2 | DeepSeek R1 | 87.75 | #251 |
| 3 | Gemini 2.0 Flash | 86.58 | #342 |
| 4 | O1 Preview | 86.58 | #193 |
| 5 | Claude 3.5 Sonnet (20240620) | 85.96 | #299 |
| 6 | GPT-4o (2024-05-13) | 85.29 | #325 |
| 7 | Llama 3.1 405B Instruct | 84.85 | #457 |
| 8 | GPT-4 Turbo (Preview) | 83.19 | #397 |
| 9 | Mistral Large 2 (Jul) | 82.81 | #484 |
| 10 | GPT-4o | 82.52 | #343 |
| 11 | DeepSeek V3 | 82.34 | #321 |
| 12 | Llama 3.2 90B Vision Instruct | 82.07 | #622 |
| 13 | Llama 3 70B Instruct | 81.17 | #634 |
| 14 | GPT-4o (2024-08-06) | 80.17 | #336 |
| 15 | Claude 3 Opus | 80.12 | #571 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-instruction-following · How It Works · Data refreshed daily, snapshot 2026-10-20.