InternVL3.5-8B: benchmark results
OpenGVLab's 8B InternVL3.5 vision-language model (August 2025) with a Qwen3 backbone and Cascade RL training for multimodal reasoning. Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Unified ELO 1556 ± 4, rank #581 of 1607 rated models, from 723 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ArtECulture - Chinese | 54.34 | Accuracy (%; Chinese-speaking culture labels; zero-shot pred | 100 |
| CapRiCorn-1K-V - Referential Consistency | 5.4 | Subject referential consistency (%; share of same-subject de | 100 |
| Holtercare-Bench - Diagnosis (Video) | 42.13 | Accuracy (%; multiple-choice closed QA on the overall diagno | 100 |
| Holtercare-Bench - Presence (Video) | 67.97 | Accuracy (%; multiple-choice closed QA on whether a given rh | 100 |
| MMBU - Grounded Classification (Boxes) | 43.9 | Closed-ended classification of a region marked by a bounding | 100 |
| MPCI-Bench - Seed Tier Probing | 97.8 | Probe Accuracy (%) | 100 |
| MPCI-Bench - Story Tier Probing | 95.1 | Probe Accuracy (%) | 100 |
| RoboProcessBench - Phase Recognition | 37.4 | Accuracy (%) on 1,274 questions asking which coarse process | 100 |
| MMBU - Ungrounded Classification (Open-Ended) | 11.1 | Open-ended (free-form) ungrounded classification of the whol | 93.8 |
| RoboProcessBench - Current Primitive Recognition | 36.8 | Accuracy (%) on 359 questions asking which low-level primiti | 92.3 |
| SIS-Bench - Action Recognition | 56.1 | Accuracy (%; 686 action recognition questions; four-option m | 92 |
| AgroVG | 39.63 | All@.5 (self-reported) | 91.7 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-8b · How It Works · Data refreshed daily, snapshot 2026-09-29.