InternVL3-8B-Instruct: benchmark results
Provider: Shanghai AI Lab. Access: Open.
Aggregate Rating 1481 ± 24, rank #2656 of 10916 rated models, from 49 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FewMMBench (4-Shot, Random) | 79.9 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 88 |
| FewMMBench (4-Shot, Similar) | 79.9 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 88 |
| FewMMBench (8-Shot, Random) | 80.1 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 88 |
| FewMMBench (8-Shot, Similar) | 79 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 88 |
| FewMMBench (4-Shot, Similar + CoT) | 75.2 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 76 |
| FewMMBench (8-Shot, Similar + CoT) | 74.2 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 76 |
| MCSBench - L3 | 25.19 | Accuracy (%) | 65.1 |
| AICA-Bench - Emotion-Guided Generation | 67.27 | Emotion-guided content generation score on a percent scale: | 63.6 |
| MERA Multi - ruMathVQA | 3.2 | Score (%) | 57.6 |
| FewMMBench (Zero-Shot) | 72.6 | Mean accuracy (%) on the FewMMBench query sets (about 250 mu | 56 |
| AICA-Bench | 62.35 | Overall average on a percent scale: unweighted mean of the e | 50 |
| AICA-Bench - Emotion Understanding (CoT) | 52.98 | Weighted F1 (%) of emotion classification (expressed emotion | 45.5 |
Interactive version: theaggregate.ai/model?slug=internvl3-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.