LLaVA-NeXT-Video-DPO (34B): benchmark results

Provider: LLaVA Team.

Aggregate Rating 1489 ± 52, rank #2485 of 10556 rated models, from 17 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
VideoHallucer32.3Overall pair accuracy over five hallucination settings (%)73.3
VideoOdyssey-V - Temporal Ordering36Multiple-choice accuracy (%) on the 89 temporal ordering que55.9
VideoOdyssey-V - Attribute Recognition34Multiple-choice accuracy (%) on the 106 attribute recognitio38.2
VideoOdyssey-V - Action Recognition26.6Multiple-choice accuracy (%) on the 192 action recognition q29.4
VideoOdyssey-V - Captioning34.7Multiple-choice accuracy (%) on the 49 captioning questions 29.4
VideoOdyssey-V - Intentional Reasoning33.3Multiple-choice accuracy (%) on the 66 intentional reasoning29.4
VideoOdyssey-V - Summarization30.8Multiple-choice accuracy (%) on the 65 summarization questio23.5
VideoOdyssey-V - Emotional Reasoning31Multiple-choice accuracy (%) on the 71 emotional reasoning q20.6
VideoOdyssey-V - OCR27.5Multiple-choice accuracy (%) on the 167 OCR questions (recog17.6
VideoOdyssey-V - Spatial Reasoning24.6Multiple-choice accuracy (%) on the 61 spatial reasoning que17.6
VideoOdyssey-V - Temporal Grounding26.2Multiple-choice accuracy (%) on the 65 temporal grounding qu17.6
VideoOdyssey-V25.7Multiple-choice accuracy (%) over all questions of VideoOdys14.7

Interactive version: theaggregate.ai/model?slug=llava-next-video-dpo-34b · How It Works · Data refreshed daily, snapshot 2026-11-17.