Ovis2-4B: benchmark results
Provider: Alibaba. Released 2025-02-21. Access: Open.
Aggregate Rating 1497 ± 19, rank #2409 of 10916 rated models, from 536 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Body Emotion Recognition | 80 | Score (%) | 100 |
| OpenVLM MMT-Bench - PLP | 56.2 | Score (%) | 100 |
| OpenVLM MMT-Bench - Fashion Recognition | 90 | Score (%) | 99 |
| OpenVLM MMT-Bench - AUD | 69 | Score (%) | 98.8 |
| OpenVLM MMT-Bench - Color Contrast | 85 | Score (%) | 98.8 |
| OpenVLM Q-Bench - What (Distortion) | 82.7 | Accuracy (%) | 98.7 |
| OpenVLM MMT-Bench - IEJ | 65 | Score (%) | 98.3 |
| OpenVLM OCRBench - Text Recognition | 95 | Correct Answers (%) | 98.2 |
| OpenVLM MMT-Bench - Pixel Localization | 55 | Score (%) | 98.1 |
| OpenVLM MTVQA - Korean | 37.5 | Accuracy (%) | 98.1 |
| OpenVLM MMBench V1.1 CN - Identity Reasoning | 100 | Accuracy (%) | 97.9 |
| OpenVLM OCRBench - Key Information Extraction | 93.5 | Correct Answers (%) | 97.9 |
Interactive version: theaggregate.ai/model?slug=ovis2-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.