HELM Capabilities - WildBench: leaderboard
HELM Capabilities run of WildBench, scoring response quality on challenging real-user instructions.
Metric: WB Score. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 (2025-11-13) | 86.29 | #101 |
| 2 | Kimi K2 | 86.19 | #243 |
| 3 | O3 (2025-04-16) | 86.07 | #121 |
| 4 | Gemini 3 Pro (Preview) | 85.87 | #66 |
| 5 | GPT-5 | 85.74 | #98 |
| 6 | Gemini 2.5 Pro (Preview 03-25) | 85.67 | #123 |
| 7 | GPT-5 Mini (2025-08-07) | 85.53 | #172 |
| 8 | Claude Sonnet 4.5 | 85.44 | #142 |
| 9 | GPT-4.1 (2025-04-14) | 85.43 | #250 |
| 10 | O4 Mini (2025-04-16) | 85.41 | #179 |
| 11 | Claude Opus 4 (20250514) (Thinking) | 85.19 | #154 (Claude Opus 4 (20250514)) |
| 12 | Grok 3 | 84.94 | #304 |
| 13 | GPT-OSS-120B | 84.49 | #341 |
| 14 | Claude Haiku 4.5 (20251001) | 83.91 | #261 |
| 15 | Claude Sonnet 4 (20250514) (Thinking) | 83.81 | #213 (Claude Sonnet 4 (20250514)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-capabilities-wildbench · How It Works · Data refreshed daily, snapshot 2026-10-20.