HELM VHELM - Exams V - Bulgarian (Natural Science): leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) Exams-V questions in Bulgarian on natural science: school exam multiple-choice questions rendered as images that combine text, formulas and figures; 54 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 | GPT-4.1 (2025-04-14) | 46.3 | #257 |
| 2 | GPT-4.5 | 40.74 | #278 |
| 3 | GPT-4.1 Mini | 33.33 | #363 |
| 4 | GPT-4o (2024-08-06) | 33.33 | #343 |
| 5 | Gemini 2.5 Pro (Preview 03-25) | 31.48 | #126 |
| 6 | Llama 4 Maverick Instruct FP8 | 22.22 | #424 |
| 7 | GPT-4o (2024-05-13) | 20.37 | #337 |
| 8 | Claude 3.5 Sonnet (20241022) | 14.81 | #302 |
| 9 | Gemini 2.0 Flash | 12.96 | #351 |
| 10 | Gemini 1.5 Pro (002) | 11.11 | #368 |
| 11 | Claude 3.5 Sonnet (20240620) | 11.11 | #308 |
| 12 | GPT-4o (2024-11-20) | 9.26 | #392 |
| 13 | Gemini 2.0 Flash (Preview) | 7.41 | #376 |
| 14 | Gemini 2.0 Pro (Preview 02-05) | 5.56 | #318 |
| 15 | Claude 3 Opus (20240229) | 3.7 | #428 |
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-exams-v-bulgarian-natural-science · How It Works · Data refreshed daily, snapshot 2026-11-05.