InternVL3.5-8B-Instruct: benchmark results
Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Aggregate Rating 1536 ± 17, rank #1877 of 10916 rated models, from 129 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SiT-Bench - Global Perception & Mapping | 26.14 | Accuracy (%; item-weighted over the category's subtasks) | 90 |
| SiT-Bench - Embodied & Fine-grained Perception | 48.78 | Accuracy (%; item-weighted over the category's subtasks) | 86.7 |
| SiT-Bench | 43.27 | Accuracy (%; item-weighted over the 17 subtasks) | 86.2 |
| UniKIE-Bench (Open-Category) - English Receipt | 79.46 | Field-level F1 (%) on the 207 English receipt documents of U | 85.7 |
| DatBench - Spatial Reasoning | 24.1 | Accuracy (%) | 84.6 |
| Video-MME-Logical | 12.1 | Overall accuracy (%) over all test videos; Video-MME-Logical | 81.2 |
| SiT-Bench - Logic & Anomaly Detection | 37.78 | Accuracy (%; item-weighted over the category's subtasks) | 80 |
| SiT-Bench - Navigation & Planning | 49.78 | Accuracy (%; item-weighted over the category's subtasks) | 80 |
| DatBench - Diagrams & Tables | 44.5 | Accuracy (%) | 76.9 |
| MIOH - Existence | 60.4 | Accuracy (%; existence questions about whether objects are p | 74.1 |
| MIOH | 36.9 | Accuracy (%; mean of the existence, counting, attribute and | 70.4 |
| SpatiaLQA - Precondition F1 | 19.2 | Precondition F1 (%, 0-100): harmonic mean of the pooled reca | 69.2 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.