InternVL2.5-38B-MPO: benchmark results
Provider: Shanghai AI Lab. Released 2024-12-20. Access: Open.
Aggregate Rating 1570 ± 18, rank #1397 of 10916 rated models, from 559 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM A-Bench - Aesthetic Quality | 72.1 | Accuracy (%) | 100 |
| OpenVLM MMMU - Management | 83.3 | Accuracy (%) | 99.6 |
| OpenVLM AI2D - Solar System | 100 | Accuracy (%) | 99.3 |
| OpenVLM MMT-Bench - Spot The Similarity | 100 | Score (%) | 99.3 |
| OpenVLM MMT-Bench - Web Shopping | 65 | Score (%) | 99.3 |
| OpenVLM OCRBench - Doc-oriented VQA | 93.5 | Correct Answers (%) | 99.3 |
| OpenVLM OCRBench - Key Information Extraction | 94 | Correct Answers (%) | 99.3 |
| OpenVLM SEEDBench IMG - Instance Identity | 82 | Accuracy (%) | 99.3 |
| OpenVLM ScienceQA Test | 99.1 | Accuracy (%) | 99.1 |
| OpenVLM MMT-Bench - Film and Television Recognition | 100 | Score (%) | 99 |
| OpenVLM OCRBench - Scene Text-centric VQA | 96.5 | Correct Answers (%) | 98.9 |
| OpenVLM MMT-Bench - Temporal Sequence Understanding | 60 | Score (%) | 98.8 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-38b-mpo · How It Works · Data refreshed daily, snapshot 2026-11-08.