InternVL2.5-78B-MPO: benchmark results
Provider: Shanghai AI Lab. Released 2024-12-20. Access: Open.
Aggregate Rating 1582 ± 20, rank #1228 of 10916 rated models, from 524 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CCBench | 86.9 | Accuracy (%) | 100 |
| OpenVLM CCBench - Sketch Reasoning | 97.8 | Accuracy (%) | 100 |
| OpenVLM MMT-Bench - Google Apps | 80 | Score (%) | 100 |
| OpenVLM OCRBench - Key Information Extraction | 95 | Correct Answers (%) | 100 |
| OpenVLM ScienceQA Test | 99.5 | Accuracy (%) | 100 |
| OpenVLM AI2D - Faults and Earthquakes | 96.4 | Accuracy (%) | 99.8 |
| OpenVLM A-Bench - Major Object Recognition | 96.2 | Accuracy (%) | 99.7 |
| OpenVLM LLaVA-Bench - Conv | 103.9 | Score | 99.6 |
| OpenVLM MMT-Bench - IR | 85.7 | Score (%) | 99.5 |
| OpenVLM MMT-Bench - VPU | 95 | Score (%) | 99.5 |
| OpenVLM OCRBench - Scene Text-centric VQA | 97.5 | Correct Answers (%) | 99.5 |
| OpenVLM ScienceQA Val | 99 | Accuracy (%) | 99.5 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-78b-mpo · How It Works · Data refreshed daily, snapshot 2026-11-08.