Multi-IF: leaderboard
4,501 three-turn instruction-following conversations in 8 languages built from IFEval prompts (Meta, 2024); each turn adds verifiable constraints; average accuracy across turns and languages.
Metric: Average (imported). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Around July 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ling-3.0-flash | 87.7 | #206 |
| 2 | O1 Preview | 87.7 | #193 |
| 3 | Llama 3.1 405B | 85.4 | #466 |
| 4 | O1 Mini | 85.3 | #323 |
| 5 | GPT-4o | 84.3 | #343 |
| 6 | Qwen 2.5 72B | 83.7 | #499 |
| 7 | Llama 3.1 70B | 82.6 | #592 |
| 8 | Nemotron 3.5 Lightning | 82.34 | #467 |
| 9 | Gemma 4 26B A4B | 82.19 | #358 |
| 10 | Claude 3.5 Sonnet | 81.7 | #350 |
| 11 | Qwen 3.6 35B A3B | 81.53 | #226 |
| 12 | GPT-4 | 81.5 | #438 |
| 13 | Mistral Large 2 (Jul) | 80.5 | #484 |
| 14 | Claude 3 Sonnet | 78.2 | #775 |
| 15 | Gemini 1.5 Pro | 75.8 | #445 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=multi-if · How It Works · Data refreshed daily, snapshot 2026-10-20.