AgentIF: leaderboard
Agent instruction-following benchmark measuring constraint and instruction success across vanilla, condition, example, formatting, semantic, and tool constraints.
Metric: Constraint Success Rate (imported). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 Mini | 59.8 | #323 |
| 2 | GPT-4o | 58.5 | #343 |
| 3 | Qwen 3 32B | 58.4 | #443 |
| 4 | QwQ-32B | 58.1 | #421 |
| 5 | DeepSeek R1 | 57.9 | #251 |
| 6 | DeepSeek V3 | 56.7 | #321 |
| 7 | Claude 3.5 Sonnet | 56.6 | #350 |
| 8 | Llama 3.1 70B Instruct | 56.3 | #561 |
| 9 | DeepSeek R1 Distill Qwen 32B | 55.1 | #655 |
| 10 | DeepSeek R1 Distill Llama 70B | 55 | #594 |
| 11 | Llama 3.1 8B Instruct | 53.6 | #1035 |
| 12 | Mistral 7B Instruct (v0.3) | 46.8 | #1274 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=agentif · How It Works · Data refreshed daily, snapshot 2026-10-20.