LLaVA-OneVision-2-8B: benchmark results
Provider: LLaVA Team. Released 2026-04-30. Access: Open.
Aggregate Rating 1607 ± 36, rank #945 of 10916 rated models, from 26 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| NAMEBench - Asserted Object Hallucination | 5.7 | Hallucination Rate (%) | 100 |
| VSTAT - Location | 43 | Score (%) on questions about location-type state elements: 1 | 100 |
| VSTAT | 35.1 | Average score (%) over all questions: 1,500 questions on 834 | 81.8 |
| VSTAT - Attribute | 40.5 | Score (%) on questions about attribute-type state elements: | 81.8 |
| VSTAT - Set | 46.9 | Score (%) on questions about set state structures: 1,500 que | 81.8 |
| VSTAT - Atomic | 33.5 | Score (%) on questions about atomic state structures: 1,500 | 77.3 |
| VSTAT - Sequence | 38.7 | Score (%) on questions about sequence state structures: 1,50 | 77.3 |
| EgoActionBench - Semantic Verb | 25.8 | Verb agreement judged by the language-model judge (0-100; al | 66.7 |
| EgoActionBench - EPIC-KITCHENS-100 subset | 28.2 | EgoActionScore (0-100; mean of Temporal Overall and Semantic | 61.1 |
| EgoActionBench - Ego4D subset | 23.9 | EgoActionScore (0-100; mean of Temporal Overall and Semantic | 61.1 |
| EgoActionBench - Semantic Objects | 32.6 | Objects agreement judged by the language-model judge (0-100; | 61.1 |
| EgoActionBench - Semantic Tools | 48.1 | Tools agreement judged by the language-model judge (0-100; a | 61.1 |
Interactive version: theaggregate.ai/model?slug=llava-onevision-2-8b · How It Works · Data refreshed daily, snapshot 2026-11-08.