InternVL2.5-4B: benchmark results
Provider: Shanghai AI Lab. Released 2024-12-05. Access: Open.
Aggregate Rating 1486 ± 19, rank #2584 of 10916 rated models, from 187 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMBench V1.1 EN - Image Emotion | 83.3 | Accuracy (%) | 94.9 |
| OpenVLM MMBench V1.1 CN - Image Scene | 91.7 | Accuracy (%) | 93.8 |
| OpenVLM MMBench V1.1 CN - Image Emotion | 85.6 | Accuracy (%) | 92.2 |
| OpenVLM MMBench V1.1 EN - Image Style | 84.8 | Accuracy (%) | 91.9 |
| OpenVLM MMBench V1.1 CN - Social Relation | 89 | Accuracy (%) | 91.5 |
| OpenVLM MMMU - Sociology | 73.3 | Accuracy (%) | 91.3 |
| OpenVLM MMBench V1.1 EN - Social Relation | 90.1 | Accuracy (%) | 90.3 |
| OpenVLM MMMU - History | 80 | Accuracy (%) | 89.8 |
| OpenVLM MMMU - Energy and Power | 56.7 | Accuracy (%) | 89.4 |
| DatBench - Spatial Reasoning | 25.4 | Accuracy (%) | 88.5 |
| OpenVLM OCRBench - Handwritten Mathematical Expression Recognition | 76 | Correct Answers (%) | 87.3 |
| OpenVLM MMMU - Humanities and Social Science | 76.7 | Accuracy (%) | 86.7 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.