InternVL2-4B: benchmark results
Provider: Shanghai AI Lab. Released 2024-07-04. Access: Open.
Aggregate Rating 1434 ± 24, rank #4439 of 10916 rated models, from 473 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GeoHeight-Bench - Threshold Segmentation | 46.44 | Text accuracy (%) on the Threshold Segmentation task (the mo | 100 |
| GeoHeight-Bench+ | 17.5 | Text accuracy (%), the mean of the 8 task accuracies, on Geo | 100 |
| GeoHeight-Bench+ - Pixel Information | 16.58 | Text accuracy (%) on the Pixel Information task (the model g | 100 |
| OpenVLM MMT-Bench - Disease Diagnosis | 100 | Score (%) | 99.8 |
| OpenVLM MMT-Bench - Multiple View Image Understanding | 50 | Score (%) | 98.1 |
| OpenVLM MMT-Bench - Handwritten Mathematical Expression Recognition | 100 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - Image Matting | 40 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - MemU | 90 | Score (%) | 97.6 |
| OpenVLM MMT-Bench - Disaster Recognition | 100 | Score (%) | 96.8 |
| OpenVLM MMMU - Design | 86.7 | Accuracy (%) | 95.8 |
| OpenVLM MMT-Bench - Meme Video Understanding | 85 | Score (%) | 95.4 |
| OpenVLM MMBench V1.1 EN - Image Topic | 98.9 | Accuracy (%) | 95.2 |
Interactive version: theaggregate.ai/model?slug=internvl2-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.