VideoZeroBench - Level-4 (Answer with 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. Level-4 accuracy: a question counts only when the model answers it correctly and its predicted evidence intervals reach a temporal IoU above 0.3; in percent of all questions, higher is better.
Metric: Level-4 accuracy (%), the share of all 500 questions answered correctly at Level-3 whose predicted evidence intervals reach temporal IoU above 0.3, 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 2029. 17 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 8 | #146 |
| 2 | Gemini 2.5 Pro | 7.2 | #280 |
| 3 | Gemini 2.5 Flash | 5.4 | #466 |
| 4 | Seed 2.0 Pro | 3.8 | #226 |
| 5 | Qwen 3 VL 235B A22B | 3.4 | #486 |
| 6 | GPT-5.2 | 1.8 | #190 |
| 7 | Qwen 3 VL 4B | 1.4 | #1163 |
| 8 | Qwen 3 VL 8B | 0.6 | #927 |
| 9 | Open-o3-Video-4B | 0.4 | |
| 10 | InternVL3.5-8B | 0.2 | #1618 |
| 11 | Qwen 3.5 397B A17B | 0 | #270 |
| 12 | Qwen 2.5 VL 7B | 0 | #1878 |
| 13 | InternVL3.5-4B | 0 | #1946 |
| 14 | Video-R1 | 0 | #1384 |
| 15 | VideoChat-R1.5-7B | 0 | #1260 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=videozerobench-level-4-answer-with-temporal-grounding · How It Works · Data refreshed daily, snapshot 2026-11-22.