IFEval Leaderboard: leaderboard
Instruction Following Evaluation: tests LLMs on strict instruction adherence across prompt-level and instruction-level constraints using 25+ verifiable instruction types.
Metric: Final Score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3 70B | 83.31 | #631 |
| 2 | Qwen 2 72B | 80.56 | #521 |
| 3 | Llama 3 8B | 78.84 | #1242 |
| 4 | Command-R+ | 77.01 | #891 |
| 5 | Mixtral 8x22B | 73.08 | #766 |
| 6 | Command R | 69.71 | #1223 |
| 7 | Mistral-7B-v0.1 | 60.96 | #1326 |
| 8 | Codestral-22B-v0.1 | 60.74 | #911 |
| 9 | Mixtral 8x7B (v0.1) | 60.12 | #1083 |
| 10 | Mistral-7B-v0.3 | 56.89 | #1355 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ifeval-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.