ProactiveVideoQA AURA Study - VAD Omega 0.5: leaderboard

The AURA 2.0 paper compares released models on streaming video interaction. IPIBench, StreamArena, ProactiveVideoQA and PhoStream retain their own protocols. ProactiveVideoQA reports PAUC; PhoStream’s overall score is question-weighted, and its timing categories are not guarantees of answer correctness. This board reports pAUC (native). Higher is better.

Metric: PAUC (native). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-4.1 Mini47.7#781
2GPT-4.146.2#500
3MMDuet (Harness: remove assistant turns)42.5
4Gemini 1.5 Pro36.1#1061
5Gemini 2.0 Flash35.8#717
6MOSS-VL-Realtime35.3
7JoyAI-VL-Interaction31.3
8LongVA-7B29.8#2788
9Qwen 2.5 VL 7B29.3#1838
10MMDuet27.4
11InternLM-XComposer2.526.8#3981
12LLaVA-OneVision-7B25.6#2685
13VideoLLM-online-8B25
14InternVL2.5-8B21.5#2444

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=proactivevideoqa-aura-study-vad-omega-0-5 · How It Works · Data refreshed daily, snapshot 2026-11-16.