OmniPro - Target Grounding: leaderboard

OmniPro (2026) evaluates proactive streaming video understanding: from continuous audio-visual streams of 2,700 human-verified questions, a model must decide when to respond and what to say across nine sub-tasks. In probe mode the model is queried just before and just after each trigger and scores only when both answers are right; this board is the probe accuracy in percent on giving the coordinates of a specified target when it appears, higher is better.

Metric: Probe-mode accuracy (%): for each ground-truth trigger the model, given the video so far, is queried 2-5 s before (expecting a negative answer) and 0-3 s after (expecting the task answer, exact match on a structured output), and the trigger counts only if both are right, on the explicit target grounding (give the coordinates of a specified target when it appears) sub-task of OmniPro (2,700 human-verified proactive streaming questions on LongVALE and COIN videos, 84% needing audio; 1 fps input); higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1video-SALMONN-2+ 7B18.1#2322
2MiniCPM-o-4.516.4#1252
3Qwen 3 Omni 30B A3B10.4#847
4Qwen2.5-Omni-7B8.5#1884
5Phi-4 Multimodal Instruct5.1#4154
6Qwen 3 VL 8B2.8#917
7InternVL3.5-8B2.4#1604
8VideoLLaMA2.1-7B-AV1.5#4204

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=omnipro-target-grounding · How It Works · Data refreshed daily, snapshot 2026-11-21.