InternVL3.5-30B-A3B: benchmark results
Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Aggregate Rating 1578 ± 10, rank #1286 of 10916 rated models, from 239 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CLBench-V | 28.47 | Score (%; instance-weighted mean of the dataset scores, each | 100 |
| CLBench-V - Grounding (L0) | 30.8 | Score (%; instance-weighted mean of the dataset scores, each | 100 |
| CLBench-V - Knowledge Learning (L2) | 35.36 | Score (%; instance-weighted mean of the dataset scores, each | 100 |
| RetinaGroundBench - Grounding AggIoU | 12.6 | AggIoU (%) | 96.3 |
| RetinaGroundBench - Grounding F1 | 23.7 | F1 (%) | 96.3 |
| RetinaGroundBench - Grounding Recall | 16.5 | Recall (%) | 96.3 |
| SiT-Bench - Multi-View & Geometric Reasoning | 53.59 | Accuracy (%; item-weighted over the category's subtasks) | 93.3 |
| VGenST-Bench - Behavioral Trigger Identification | 90.4 | Behavioral Trigger Identification (environmental scale, exoc | 92.9 |
| RetinaGroundBench - Grounding Precision | 47.9 | Precision (%) | 92.6 |
| UNICBench (Image) | 19.1 | Exact-count hit rate (%): share of parsed numeric answers eq | 90 |
| UniSense Study - Image - MME | 2461.9 | MME Score | 85.7 |
| TabVerse - QA (LaTeX Image) | 50 | Exact-match accuracy (%) answering table questions from a re | 83.3 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-30b-a3b · How It Works · Data refreshed daily, snapshot 2026-11-08.