OVO-S-Bench: leaderboard

Metric: Overall multiple-choice accuracy (%) on the 1,680-question evaluation set, prefix-access mode (128 frames sampled uniformly from the video before the query timestamp, question given with the frames), macro-average of the four levels. Source: arxiv.org. Saturation forecast: Around 2031. 31 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)59.19
2Qwen 3 VL 235B A22B53.64
3Qwen 3.5 397B A17B52.1
4Qwen 3.5 27B (Non-reasoning)51.69
5GPT-5.450.89
6Gemini 3.1 Flash Lite50.81
7Qwen 3 VL 32B Instruct48.8
8Qwen 3 VL 4B Instruct46.83
9Qwen 3.5 9B (Non-reasoning)45.91
10Qwen 3.5 4B (Non-reasoning)45.4
11InternVL3.5-8B44.57
12Qwen 2.5 VL 7B Instruct44.19
13Grok 4.1 Fast (Non-reasoning)43.73
14Gemma 4 26B A4B42.56
15Gemma 4 E4B39.69

Interactive version: theaggregate.ai/benchmark?slug=ovo-s-bench · How It Works · Data refreshed daily, snapshot 2026-09-29.