HELM VHELM - Blink Perception - Semantic Correspondence: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) BLINK Semantic Correspondence task: matching semantically corresponding points across images, one of BLINK's perception tasks that humans solve at a glance, asked as multiple choice; 139 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 | O3 (2025-04-16) | 83.45 | #125 |
| 2 | O1 (2024-12-17) | 78.42 | #157 |
| 3 | O4 Mini (2025-04-16) | 77.7 | #183 |
| 4 | GPT-4.5 | 76.98 | #278 |
| 5 | GPT-4.1 (2025-04-14) | 74.1 | #257 |
| 6 | Gemini 2.0 Flash Lite | 74.1 | #455 |
| 7 | Gemini 2.5 Pro (Preview 03-25) | 73.38 | #126 |
| 8 | Gemini 2.0 Flash (Preview) | 73.38 | #376 |
| 9 | Gemini 1.5 Pro (002) | 71.22 | #368 |
| 10 | Gemini 2.0 Pro (Preview 02-05) | 71.22 | #318 |
| 11 | Gemini 2.0 Flash | 68.35 | #351 |
| 12 | GPT-4o (2024-08-06) | 68.35 | #343 |
| 13 | GPT-4o (2024-05-13) | 66.91 | #337 |
| 14 | Llama 4 Maverick Instruct FP8 | 66.91 | #424 |
| 15 | Claude 3.5 Sonnet (20240620) | 65.47 | #308 |
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-perception-semantic-correspondence · How It Works · Data refreshed daily, snapshot 2026-11-05.