VMMU: leaderboard
Vietnamese multimodal exam benchmark: 2,548 multiple-choice questions from the national high-school graduation exam (math, physics, chemistry, biology, geography), the driving-licence test and IQ tests, each shown to a vision-language model as one rendered page image. Scored by mean accuracy over the seven domains, higher is better (Dang et al., AACL-IJCNLP 2026 Findings).
Metric: Mean accuracy (%) over the seven domains of 2,548 Vietnamese multiple-choice exam questions, each given as one rendered page image with the question, options and visual evidence (answer letter extracted from curly brackets; random baseline 25.9); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 86.33 | #137 |
| 2 | GPT-5 (Medium) | 78.41 | #167 (GPT-5) |
| 3 | Claude Opus 4.6 (Thinking) | 74.62 | #102 (Claude Opus 4.6) |
| 4 | O3 (Medium) | 73.98 | #228 (O3) |
| 5 | Gemini 2.5 Flash | 71.34 | #453 |
| 6 | Qwen 2.5 VL 32B Instruct | 52.13 | #1053 |
| 7 | Qwen 2.5 VL 72B Instruct | 51.75 | #846 |
| 8 | GPT-4.1 | 51.58 | #495 |
| 9 | Claude Sonnet 4 | 50.58 | #400 |
| 10 | Llama 4 Scout | 49.06 | #1776 |
| 11 | Gemma 3 27B | 39.09 | #1631 |
| 12 | Llama 4 Maverick | 38.37 | #1099 |
| 13 | Mistral Medium 3 | 37.78 | #1073 |
| 14 | Mistral Small 3.2 | 31.88 | #1591 |
| 15 | Gemma 3 4B | 25.45 | #5064 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vmmu · How It Works · Data refreshed daily, snapshot 2026-11-13.