OpenVLM MTVQA - Thai: leaderboard
MTVQA Thai accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): questions about Thai text in scene and document images, asked and answered in Thai. One of the nine language splits of the multilingual text-centric VQA benchmark; accuracy in percent, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 157 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 33.3 | #363 |
| 2 | GPT-4.1 (2025-04-14) | 29.4 | #257 |
| 3 | Gemini 1.5 Pro | 26.8 | #454 |
| 4 | InternVL3-14B | 24.2 | #506 |
| 5 | Claude 3.7 Sonnet | 23.4 | #259 |
| 6 | InternVL3-78B | 22.9 | #361 |
| 7 | GPT-4.1 Nano | 22.1 | #730 |
| 8 | Gemini 1.5 Flash | 22.1 | #620 |
| 9 | InternVL3-38B | 21.2 | #413 |
| 10 | Gemma 3 27B | 20.3 | #606 |
| 11 | InternVL3-8B | 18.2 | #624 |
| 12 | Grok 2 (1212) | 17.3 | #502 |
| 13 | Gemma 3 12B | 16.9 | #674 |
| 14 | Gemma 3 4B | 15.6 | #1088 |
| 15 | Qwen 2.5 VL 7B | 15.2 | #641 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-mtvqa-thai · How It Works · Data refreshed daily, snapshot 2026-11-05.