InternVL3.5-30B-A3B-Instruct: benchmark results
Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Aggregate Rating 1577 ± 26, rank #1300 of 10916 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Video-IFBench - Nested | 9.6 | Task-gated instruction satisfaction rate (%; TISR on nested | 67.6 |
| Video-MME-v2 - Level 1 Retrieval & Aggregation (w/o subs) | 17.48 | Level 1 Score w/o sub (%) | 64.6 |
| GAVEL - Grounding Accuracy (IoU 0.5) | 15.8 | Accuracy (%; share of predicted evidence boxes with IoU of a | 60 |
| Video-MME-Logical | 8.7 | Overall accuracy (%) over all test videos; Video-MME-Logical | 56.2 |
| CulMind - Ancient Scripts | 5.3 | Ancient Scripts subdomain score: macro-average of the task s | 53.8 |
| Video-MME-v2 | 28.7 | Avg Accuracy w/o sub (%) | 50 |
| Video-MME-v2 - Level 3 Complex Reasoning (w/o subs) | 10.94 | Level 3 Score w/o sub (%) | 50 |
| Video-MME-v2 - Non-Lin Score (w/o subs) | 12.6 | Non-Lin Score w/o sub (%) | 49 |
| Video-IFBench - Selection | 13.8 | Task-gated instruction satisfaction rate (%; TISR on selecti | 40.5 |
| CulMind - Ancient Books | 37.8 | Ancient Books subdomain score: macro-average of the task sco | 38.5 |
| Video-MME-v2 - Level 2 Temporal Understanding (w/o subs) | 11.04 | Level 2 Score w/o sub (%) | 33.3 |
| Video-IFBench - Multi | 24.8 | Task-gated instruction satisfaction rate (%; TISR on multi-t | 28.4 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-30b-a3b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.