OmniVinci: benchmark results
Provider: NVIDIA. Released 2025-10-19. Access: Open.
Aggregate Rating 1532 ± 33, rank #1930 of 10916 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| VI-Bench (Video Prompt Inversion) - Hard (Multi-Shot) | 0.45 | Inversion Score (0-1; mean of the Prompt Score, GPT-4o (gpt- | 93.8 |
| VI-Bench (Video Prompt Inversion) | 0.58 | Inversion Score (0-1; mean of the Prompt Score, GPT-4o (gpt- | 86.7 |
| SpeakerSleuth - Detection | 57.5 | Balanced accuracy (%; 0.5 x (S1 + mean(S2, S3)) over 606 dia | 81.8 |
| IMAVB (Fixed Option Order) | 38.4 | Balanced accuracy (%; mean of the four split accuracies, 2,0 | 75 |
| VI-Bench (Video Prompt Inversion) - Easy (Semantic Grounding) | 0.72 | Inversion Score (0-1; mean of the Prompt Score, GPT-4o (gpt- | 75 |
| SpeakerSleuth - Discrimination Ranking | 25.4 | Exact match of the full three-candidate ranking (%; 606 item | 72.7 |
| VI-Bench (Video Prompt Inversion) - Medium (Style and Camera) | 0.56 | Inversion Score (0-1; mean of the Prompt Score, GPT-4o (gpt- | 65.6 |
| SpeakerSleuth - Discrimination | 45.4 | Classification accuracy (%; 606 three-candidate items) | 54.5 |
| IMAVB | 36.3 | Balanced accuracy (%; mean of standard vision, standard audi | 37.5 |
| CapRiCorn-1K - Referential Consistency | 1.2 | Subject referential consistency (%; share of same-subject de | 23.5 |
| DiaDemBench - Speaker attribution F1 (REF) | 17.1 | REF F1 | 23.1 |
| DiaDemBench - Utterance transcription F1 (ASR) | 25.2 | ASR F1 | 23.1 |
Interactive version: theaggregate.ai/model?slug=omnivinci · How It Works · Data refreshed daily, snapshot 2026-11-08.