HELM VHELM - Unicorn - Sketchy-VQA: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) Unicorn Sketchy-VQA split, an out-of-distribution safety test: questions about sketch drawings of objects; 1,000 questions, prefix quasi-exact match. Percent correct, 0-100, higher is better.
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 97 | #351 |
| 2 | Gemini 2.0 Pro (Preview 02-05) | 97 | #318 |
| 3 | Gemini 2.0 Flash (Preview) | 96.3 | #376 |
| 4 | Gemini 1.5 Pro (002) | 96.2 | #368 |
| 5 | Gemini 2.5 Pro (Preview 03-25) | 96.2 | #126 |
| 6 | Gemini 2.0 Flash Lite | 96.2 | #455 |
| 7 | Gemini 1.5 Flash (002) | 96.1 | #587 |
| 8 | GPT-4o (2024-08-06) | 94.9 | #343 |
| 9 | GPT-4.5 | 94.7 | #278 |
| 10 | GPT-4o (2024-05-13) | 93.6 | #337 |
| 11 | Claude 3.5 Sonnet (20241022) | 92.7 | #302 |
| 12 | GPT-4o (2024-11-20) | 92.7 | #392 |
| 13 | O4 Mini (2025-04-16) | 92.5 | #183 |
| 14 | GPT-4.1 (2025-04-14) | 92.3 | #257 |
| 15 | Llama 4 Scout Instruct | 92 | #547 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-unicorn-sketchy-vqa · How It Works · Data refreshed daily, snapshot 2026-11-05.