VideoZeroBench - Temporal Grounding: leaderboard

VideoZeroBench (2026) asks 500 manually annotated, deliberately hard open-ended questions with exact short answers about 138 long videos in 13 domains, each with annotated temporal intervals and bounding boxes as supporting evidence. Mean temporal IoU between the evidence intervals the model predicts and the annotated ones, in percent; higher is better.

Metric: Mean multi-segment temporal IoU (%) between the evidence intervals the model predicts from the video and question and the annotated intervals (0 for a missing or invalid interval or a question without temporal evidence), on VideoZeroBench's 500 manually annotated open-ended questions (exact short answers) over 138 long videos in 13 domains; Gemini models get the raw video through the API, the others 1 fps frames up to the printed cap, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro32#145
2Gemini 2.5 Pro31.4#276
3Gemini 2.5 Flash27.9#463
4Seed 2.0 Pro25.3#222
5GPT-5.221.9#185
6Qwen 3 VL 235B A22B19.6#485
7Qwen 3 VL 4B13.3#1151
8Qwen 3 VL 8B10.9#917
9Open-o3-Video-4B9.7
10VideoRFT-7B3.6#1863
11Video-o3-7B3.5
12InternVL3.5-4B3.3#1930
13VideoChat-R1.5-7B3.3#1254
14Qwen 3.5 397B A17B3.1#264
15InternVL3.5-8B2.9#1604

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

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