Ovis2-8B: benchmark results
AIDC-AI's (Alibaba) Ovis2 8B vision-language model pairing an AIMv2 vision encoder with Qwen2.5-7B-Instruct, strong on OCR, charts and video (February 2025). Provider: Alibaba. Released 2025-02-21. Access: Open.
Aggregate Rating 1519 ± 11, rank #2067 of 10916 rated models, from 1138 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MEGA-Bench Task - Extract Webpage Headline | 85.7 | Task Score (%) | 100 |
| MEGA-Bench Task - GUI Chat Easy | 77.7 | Task Score (%) | 100 |
| MEGA-Bench Task - GUI Chat Hard | 70.4 | Task Score (%) | 100 |
| MEGA-Bench Task - Remaining Playback Time Calculation | 35.7 | Task Score (%) | 100 |
| OpenVLM MMT-Bench - Color Contrast | 90 | Score (%) | 100 |
| OpenVLM MMT-Bench - Traffic Participants Understanding | 75 | Score (%) | 99.3 |
| OpenVLM MMBench V1.1 CN - Function Reasoning | 96.7 | Accuracy (%) | 98.9 |
| OpenVLM MMT-Bench - AUD | 69 | Score (%) | 98.8 |
| OpenVLM MMT-Bench - Chart VQA | 85 | Score (%) | 98.8 |
| OpenVLM Q-Bench - What (Distortion) | 82.7 | Accuracy (%) | 98.7 |
| OpenVLM SEEDBench IMG - Visual Reasoning | 83.4 | Accuracy (%) | 98.4 |
| OpenVLM MMBench V1.0 CN - AR | 88.2 | Accuracy (%) | 98.3 |
Interactive version: theaggregate.ai/model?slug=ovis2-8b · How It Works · Data refreshed daily, snapshot 2026-11-08.