InternVL3.5-1B: benchmark results
Provider: Shanghai AI Lab. Released 2025-08-26. Access: Open.
Aggregate Rating 1399 ± 15, rank #6175 of 10916 rated models, from 189 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EgoCoT-Bench - Hand-Object Temporal Retrospection | 75 | Answer accuracy (%, four-way multiple choice, strict exact m | 97.2 |
| Metric-Bench - Position | 32.3 | Mean relative accuracy (%; object-position questions; 1,340 | 85.7 |
| PRISM-VLM - Sycophancy Resistance | 72.2 | Per-axis mean score (%; share of turn-0-correct answers held | 85.4 |
| Cambrian-S Release - VSI-Bench | 49.9 | Source benchmark score (%) | 63.2 |
| MIOH - Position | 38 | Accuracy (%; position questions about object positions acros | 63 |
| Cambrian-S Release - VSI-Bench Debiased | 41.8 | Source benchmark score (%) | 61.1 |
| EgoCoT-Bench - Goal-Oriented Object Tracking | 32.55 | Answer accuracy (%, four-way multiple choice, strict exact m | 55.6 |
| Cambrian-S Release - MVBench † | 61 | Accuracy (%) | 53.3 |
| TimeSpot - Daylight Phase | 35.8 | Accuracy (%) of the predicted daylight phase (such as sunris | 42.3 |
| MirrorBench - Level 0 (Guided Mirror Perception) | -0.07 | Average score (mean of task success rate, step-wise improvem | 41.2 |
| HomeSafe-Bench | 17.24 | Weighted Safety Score (0-100) on the 438 hazardous household | 40 |
| Cambrian-S Release - LongVBench † | 53 | Accuracy (%) | 38.9 |
Interactive version: theaggregate.ai/model?slug=internvl3-5-1b · How It Works · Data refreshed daily, snapshot 2026-11-08.