TraceAV-Bench - Visual-to-Audio Deception: leaderboard

The Visual-to-Audio Deception sub-task of TraceAV-Bench (2026), in its multimodal hallucination dimension: questions that claim a sound from what is seen in long audio-visual videos. The score is accuracy in percent with full audio and video input; higher is better.

Metric: Accuracy (%) on the 230 Visual-to-Audio Deception (V2A) questions of the multimodal hallucination dimension, four-option multiple choice over long audio-visual videos (10 to 140 minutes), exact match of the selected option set, each model at its official inference setting, with the full audio and video input; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)89.57#102
2Gemini 2.5 Pro79.13#280
3Gemini 3 Flash76.52#180
4Gemini 2.0 Flash74.78#731
5Gemma 4 E4B74.35#2199
6Ming-Flash-Omni-2.071.3#1258
7MiniCPM-o-4.570.87#1251
8Qwen 3 Omni 30B A3B65.65#851
9Gemini 2.5 Flash60.87#466
10Qwen2.5-Omni-7B60.43#1888
11video-SALMONN 245.65#2695
12OmniVinci-9B42.17
13VideoLLaMA2.1-7B-AV35.22#4199
14HumanOmni-7B33.91#3241
15Baichuan-Omni-1.528.26#3189

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=traceav-bench-visual-to-audio-deception · How It Works · Data refreshed daily, snapshot 2026-11-22.