HELM VHELM - Blink Reasoning - Multi-view Reasoning: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) BLINK Multi-view Reasoning task: inferring camera motion between two views, one of BLINK's perception tasks that humans solve at a glance, asked as multiple choice; 133 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 | Claude 3.5 Sonnet (20240620) | 100 | #308 |
| 2 | Llama 4 Maverick Instruct FP8 | 100 | #424 |
| 3 | Claude 3 Haiku (20240307) | 100 | #780 |
| 4 | Claude 3 Sonnet (20240229) | 100 | #694 |
| 5 | Claude 3 Opus (20240229) | 100 | #428 |
| 6 | Gemini 2.0 Flash Lite | 100 | #455 |
| 7 | GPT-4.1 (2025-04-14) | 99.25 | #257 |
| 8 | GPT-4o Mini (2024-07-18) | 98.5 | #568 |
| 9 | GPT-4 Turbo | 96.99 | #417 |
| 10 | Llama 4 Scout Instruct | 95.49 | #547 |
| 11 | O1 (2024-12-17) | 92.48 | #157 |
| 12 | Gemini 2.5 Pro (Preview 03-25) | 83.46 | #126 |
| 13 | O3 (2025-04-16) | 68.42 | #125 |
| 14 | GPT-4o (2024-11-20) | 68.42 | #392 |
| 15 | O4 Mini (2025-04-16) | 67.67 | #183 |
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-blink-reasoning-multi-view-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.