InternVL3-38B: benchmark results
Shanghai AI Lab's open InternVL3 38B vision-language model. Provider: Shanghai AI Lab. Released 2025-04-11. Access: Open.
Aggregate Rating 1610 ± 5, rank #405 of 1656 rated models, from 1442 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MEGA-Bench Task - ARC AGI | 14.3 | Task Score (%) | 100 |
| MEGA-Bench Task - Adapted CVBench Count | 78.6 | Task Score (%) | 100 |
| MEGA-Bench Task - Code Visualization Output Understanding | 70 | Task Score (%) | 100 |
| MEGA-Bench Task - Counting Single Image | 85.7 | Task Score (%) | 100 |
| MEGA-Bench Task - Entertainment Web Game Style | 92.9 | Task Score (%) | 100 |
| MEGA-Bench Task - Functionality Matching In Different Objects | 78.6 | Task Score (%) | 100 |
| MEGA-Bench Task - Signage Navigation | 92.9 | Task Score (%) | 100 |
| MEGA-Bench Task - Star Object Interaction Video | 75 | Task Score (%) | 100 |
| MEGA-Bench Task - Video Motion Matching 3D Real | 80 | Task Score (%) | 100 |
| MEGA-Bench Task - Video Motion Matching Real 3D | 85.7 | Task Score (%) | 100 |
| MEGA-Bench Task - Visual Correspondence In Two Images | 92.9 | Task Score (%) | 100 |
| MMR-AD - MVTec 3D-AD Detection | 75.9 | Anomaly detection accuracy (%): share of MVTec 3D-AD test im | 100 |
Interactive version: theaggregate.ai/model?slug=internvl3-38b · How It Works · Data refreshed daily, snapshot 2026-10-20.