OpenVLM MTVQA - Italian: leaderboard
MTVQA Italian accuracy from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): questions about Italian text in scene and document images, asked and answered in Italian. 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 | 45.5 | #363 |
| 2 | InternVL3-38B | 44.5 | #413 |
| 3 | InternVL3-78B | 43.7 | #361 |
| 4 | GPT-4.1 (2025-04-14) | 43.3 | #257 |
| 5 | InternVL3-14B | 43.2 | #506 |
| 6 | Gemini 1.5 Pro | 42.9 | #454 |
| 7 | InternVL3-8B | 41.4 | #624 |
| 8 | Gemini 1.5 Flash | 41.4 | #620 |
| 9 | Qwen 2 VL 72B | 40.8 | #450 |
| 10 | Gemma 3 27B | 39.7 | #606 |
| 11 | Claude 3.7 Sonnet | 37.3 | #259 |
| 12 | Qwen 2.5 VL 7B | 37.2 | #641 |
| 13 | Qwen 2 VL 7B | 36.9 | #743 |
| 14 | GPT-4.1 Nano | 36.1 | #730 |
| 15 | MiniCPM-V-2.6 | 35.1 | #827 |
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-italian · How It Works · Data refreshed daily, snapshot 2026-11-05.