HELM VHELM - Mm Star Reasoning - Math: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) MM-Star Math subset: multiple-choice questions curated so that they cannot be answered without the image; 250 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) | 91.6 | #125 |
| 2 | O4 Mini (2025-04-16) | 89.2 | #183 |
| 3 | Gemini 2.5 Pro (Preview 03-25) | 87.6 | #126 |
| 4 | O1 (2024-12-17) | 87.2 | #157 |
| 5 | Gemini 2.0 Pro (Preview 02-05) | 86.4 | #318 |
| 6 | Gemini 2.0 Flash (Preview) | 83.6 | #376 |
| 7 | Gemini 1.5 Pro (002) | 81.2 | #368 |
| 8 | GPT-4.5 | 80.4 | #278 |
| 9 | Gemini 2.0 Flash | 78.8 | #351 |
| 10 | Gemini 2.0 Flash Lite | 76.4 | #455 |
| 11 | GPT-4o (2024-05-13) | 69.6 | #337 |
| 12 | GPT-4.1 (2025-04-14) | 69.2 | #257 |
| 13 | GPT-4o (2024-08-06) | 65.2 | #343 |
| 14 | Claude 3 Sonnet (20240229) | 58.4 | #694 |
| 15 | Gemini 1.5 Flash (002) | 58 | #587 |
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-mm-star-reasoning-math · How It Works · Data refreshed daily, snapshot 2026-11-05.