VMLU: leaderboard
VMLU (Vietnamese Multitask Language Understanding) benchmark — average accuracy across STEM, social-science, humanities, and other Vietnamese-language subjects.
Metric: Average (%). Source: vmlu.ai. Saturation forecast: Estimated already saturated. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | QwQ-32B | 76.13 | #421 |
| 2 | Llama 3 70B | 66.44 | #631 |
| 3 | GPT-4 | 65.53 | #438 |
| 4 | QwQ 32B-Preview | 63.9 | #651 |
| 5 | GPT-4o Mini | 62.87 | #599 |
| 6 | Gemma 2 9B (IT) | 59.04 | #788 |
| 7 | Qwen 2.5 7B Instruct | 57.51 | #862 |
| 8 | ChatGPT | 46.33 | #900 |
| 9 | gemma-7B (IT) | 41.9 | #1496 |
| 10 | Phi-3-small-128k-instruct | 41.24 | #1105 |
| 11 | Phi-3-small-8k-instruct | 40.88 | #967 |
| 12 | Qwen-7B | 32.81 | #1402 |
| 13 | Qwen 2 7B Instruct | 28.85 | #991 |
| 14 | gemma-2B (IT) | 27.72 | #1559 |
| 15 | bloom-1b7 | 25.51 | #1628 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vmlu · How It Works · Data refreshed daily, snapshot 2026-10-20.