HELM VHELM - Pairs: leaderboard
The "Pairs" subtask of HELM VHELM. Stanford HELM's VHELM project, the holistic evaluation of vision-language models: image-grounded question answering, OCR, chart and diagram reading, spatial reasoning, robustness, bias, fairness and toxicity scenarios scored under one ha...
Metric: PEM. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-11-20) | 97.22 | #378 |
| 2 | GPT-4o (2024-08-06) | 95.37 | #335 |
| 3 | GPT-4.1 Nano | 92.97 | #735 |
| 4 | O1 (2024-12-17) | 92.15 | #152 |
| 5 | GPT-4.5 | 89.81 | #276 |
| 6 | GPT-4.1 Mini | 89.19 | #351 |
| 7 | GPT-4o (2024-05-13) | 87.25 | #326 |
| 8 | Gemini 2.0 Pro (Preview 02-05) | 87.15 | #305 |
| 9 | GPT-4 Turbo | 86.07 | #406 |
| 10 | Gemini 2.0 Flash Lite | 84.75 | #453 |
| 11 | Gemini 2.5 Pro (Preview 03-25) | 83.84 | #124 |
| 12 | GPT-4o Mini (2024-07-18) | 79.59 | #561 |
| 13 | GPT-4.1 (2025-04-14) | 77.41 | #250 |
| 14 | O3 (2025-04-16) | 76.52 | #123 |
| 15 | Gemini 1.5 Pro (002) | 76.25 | #350 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-pairs · How It Works · Data refreshed daily, snapshot 2026-10-23.