OVO-S-Bench - Context Tracking: leaderboard

Metric: Level 2 (spatiotemporal context tracking of evidence no longer in view) 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), mean of its three task families. Source: arxiv.org. Saturation forecast: Around 2030. 31 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)64.04
2GPT-5.457.59
3Qwen 3.5 397B A17B55.39
4Qwen 3 VL 235B A22B55.18
5Qwen 3.5 27B (Non-reasoning)55.15
6Gemini 3.1 Flash Lite52.19
7Qwen 3 VL 32B Instruct51.91
8Qwen 3.5 9B (Non-reasoning)49.37
9Qwen 3.5 4B (Non-reasoning)48.2
10Qwen 3 VL 4B Instruct48.18
11Grok 4.1 Fast (Non-reasoning)46.59
12Gemma 4 26B A4B46.55
13InternVL3.5-8B45.83
14Qwen 2.5 VL 7B Instruct45.51
15Gemma 4 E4B42.76

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