WildBench — leaderboard
Challenging real-user instruction-following benchmark reporting WildBench scores, task-category scores, Elo, and comparison metrics.
Metric: WB Score Task-Macro. Source: huggingface.co. Status: saturated. 63 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4o (2024-05-13) | 59.3 |
| 2 | GPT-4o Mini (2024-07-18) | 57.14 |
| 3 | Mistral Large 2 (Jul) | 55.57 |
| 4 | Yi Large (Preview) | 55.29 |
| 5 | GPT-4 Turbo | 55.22 |
| 6 | Claude 3.5 Sonnet (20240620) | 54.7 |
| 7 | gemma-2-9B-it-SimPO | 53.28 |
| 8 | gemma-2-9B-it-DPO | 53.22 |
| 9 | Gemini 1.5 Pro | 52.95 |
| 10 | GPT-4 Preview (0125) | 52.28 |
| 11 | Claude 3 Opus (20240229) | 51.71 |
| 12 | Yi Large | 48.93 |
| 13 | Gemini 1.5 Flash | 48.85 |
| 14 | Gemma 2 27B (IT) | 48.54 |
| 15 | DeepSeek V2 Chat | 48.21 |
Interactive version: theaggregate.ai/benchmark?slug=wildbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.