InternVL3.5-38B: benchmark results
Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Aggregate Rating 1598 ± 7, rank #1035 of 10916 rated models, from 714 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgroVG | 39.78 | All@.5 (self-reported) | 100 |
| AgroVG - Set-F1 at IoU 0.75 | 25.69 | Macro Set-F1 at IoU 0.75 (%, times 100): query-level F1 of t | 100 |
| AgroVG - Single-Target Accuracy | 35.27 | Strict single-target accuracy (%, times 100): share of singl | 100 |
| EarthSpatialBench (Direct) - Direction Counting | 1.55 | Mean absolute error of the object count on EarthSpatialBench | 100 |
| EarthSpatialBench (Direct) - Distance Counting | 0.89 | Mean absolute error of the object count on EarthSpatialBench | 100 |
| MATEO (Text-only, ICL) | 59 | Consistent temporal-execution-order accuracy (%) on MATEO's | 100 |
| MI-CXR | 41.8 | Overall accuracy over the TEL, ICR and GTS task families, fi | 100 |
| MI-CXR - Interval-wise Change Reasoning | 57.1 | Interval-wise Change Reasoning: find the interval matching a | 100 |
| MI-CXR - Trajectory Summary (Multiple Abnormalities) | 51.5 | Global Trajectory Summarization comparing several abnormalit | 100 |
| MMR-AD - MVTec AD Detection | 84.9 | Anomaly detection accuracy (%): share of MVTec AD test image | 100 |
| MMR-AD - VisA Detection | 74.1 | Anomaly detection accuracy (%): share of VisA test images (f | 100 |
| MVX-Bench - Sequence Reasoning | 50.4 | Accuracy (%) on the 117 Sequence Reasoning questions (order | 100 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-38b · How It Works · Data refreshed daily, snapshot 2026-11-08.