HELM VHELM - Blink Perception - Counting: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) BLINK Counting task: counting objects in an image, one of BLINK's perception tasks that humans solve at a glance, asked as multiple choice; 120 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 2.5 Pro (Preview 03-25) | 85 | #126 |
| 2 | Gemini 2.0 Flash (Preview) | 81.67 | #376 |
| 3 | GPT-4.5 | 81.67 | #278 |
| 4 | O3 (2025-04-16) | 80.83 | #125 |
| 5 | O4 Mini (2025-04-16) | 80 | #183 |
| 6 | Gemini 2.0 Flash Lite | 79.17 | #455 |
| 7 | Llama 4 Scout Instruct | 78.33 | #547 |
| 8 | GPT-4 Turbo | 78.33 | #417 |
| 9 | Gemini 2.0 Pro (Preview 02-05) | 78.33 | #318 |
| 10 | Gemini 2.0 Flash | 77.5 | #351 |
| 11 | Llama 4 Maverick Instruct FP8 | 77.5 | #424 |
| 12 | GPT-4o (2024-05-13) | 76.67 | #337 |
| 13 | O1 (2024-12-17) | 76.67 | #157 |
| 14 | GPT-4o (2024-08-06) | 75.83 | #343 |
| 15 | Claude 3.5 Sonnet (20240620) | 74.17 | #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-counting · How It Works · Data refreshed daily, snapshot 2026-11-05.