VideoOdyssey-AV - Counting: leaderboard

VideoOdyssey (2026) tests ultra-long video understanding with human-written multiple-choice questions on videos averaging 109 minutes, each annotated with how long a viewer must watch continuously to answer. This board is accuracy in percent on the audio-visual subset questions about counting entities or events; higher is better.

Metric: Multiple-choice accuracy (%) on the 81 counting questions (count the occurrences of specific entities in the video based on both visual and audio cues) of VideoOdyssey-AV (human-written questions on 100 videos of 60 to 255 minutes that need both visual and audio cues; 1,062 in all), omni-modal models given video and audio, 64 frames (VITA-1.5-7B 16); higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.5 Omni Plus37#254
2Gemini 3 Flash30.9#175
3Ola-7B29.6#1557
4Gemini 3.1 Pro (Preview)28.4#99
5Unified-IO-2 XL28.4
6Gemini 2.5 Pro25.9#276
7Qwen 3 Omni 30B A3B24.7#847
8Unified-IO-2 XXL23.5
9VITA-1.5-7B23.5#1853
10VideoLLaMA2-7B16.1#2385
11Unified-IO-2 L13.6
12OneLLM12.4#2372

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

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