AnyGroundBench - Animal STVG: leaderboard
Spatio-temporal video grounding on the animal domain of AnyGroundBench (2026), Animal Kingdom and newly captured mouse-scratching videos: the model localizes the queried event in time and the target box in every frame, zero-shot. The score is vIoU@0.3, the percent of queries whose tube IoU is at least 0.3; higher is better.
Metric: vIoU@0.3 (%; share of queries whose spatio-temporal tube IoU is at least 0.3) on the 157 animal-domain test queries (Animal Kingdom and newly captured mouse-scratching videos), spatio-temporal video grounding (localize the queried event in time and the target box in every frame), zero-shot, default inference settings (Qwen3.5 with thinking disabled); higher is better. Source: arxiv.org. Saturation forecast: Around July 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 16.5 | #94 |
| 2 | Gemini 3 Flash | 14 | #164 |
| 3 | Gemini 2.5 Pro | 8.28 | #266 |
| 4 | Qwen 3.5 9B (Non-reasoning) | 4.45 | #803 (Qwen 3.5 9B) |
| 5 | GPT-5.1 | 3.18 | #229 |
| 6 | Gemini 2.5 Flash | 2.54 | #453 |
| 7 | Qwen 3.5 4B (Non-reasoning) | 2.54 | #1175 (Qwen 3.5 4B) |
| 8 | InternVL3-8B | 0.63 | #1688 |
| 9 | InternVL3-14B | 0.63 | #1182 |
| 10 | GPT-4o | 0 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=anygroundbench-animal-stvg · How It Works · Data refreshed daily, snapshot 2026-11-13.