InternVL2.5-78B: benchmark results
Shanghai AI Lab's open InternVL2.5 78B vision-language model. Provider: Shanghai AI Lab. Released 2024-12-05. Access: Open.
Aggregate Rating 1603 ± 8, rank #431 of 1656 rated models, from 938 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CG-Bench - Clue MCQ | 59.5 | Clue-Interval MCQ Accuracy (%) | 100 |
| M3-VQA (Gold Section Evidence) | 55.16 | IoU accuracy (%) between the predicted and gold answer sets | 100 |
| M3-VQA (Gold Sentence Evidence) | 58.71 | IoU accuracy (%) between the predicted and gold answer sets | 100 |
| MEGA-Bench Task - CLEVRER Physics | 50 | Task Score (%) | 100 |
| MEGA-Bench Task - CLEVRER Video Moving Object Count | 76.2 | Task Score (%) | 100 |
| MEGA-Bench Task - CVBench Adapted CVBench Count | 78.6 | Task Score (%) | 100 |
| MEGA-Bench Task - Shape Composition Colours | 54.9 | Task Score (%) | 100 |
| MEGA-Bench Task - Super CLEVR | 92.9 | Task Score (%) | 100 |
| MEGA-Bench Task - TI Fused VQA Math | 57.1 | Task Score (%) | 100 |
| MEGA-Bench Task - Video Grounding Temporal | 46.7 | Task Score (%) | 100 |
| OpenVLM MMBench V1.0 CN - CP | 90.6 | Accuracy (%) | 100 |
| OpenVLM MMBench V1.0 CN - RR | 89.1 | Accuracy (%) | 100 |
Interactive version: theaggregate.ai/model?slug=internvl2-5-78b · How It Works · Data refreshed daily, snapshot 2026-10-20.