AnyGroundBench - Sports TVG: leaderboard
Temporal video grounding on the sports domain of AnyGroundBench (2026), MultiSports and newly captured American football videos: the model gives the start and end of the queried event, zero-shot. The score is tIoU@0.3, the percent of queries whose temporal IoU with the annotation is at least 0.3; higher is better.
Metric: tIoU@0.3 (%; share of queries whose temporal IoU is at least 0.3) on the 163 sports-domain test queries (MultiSports and newly captured American football videos), temporal video grounding (the queried event's start and end), zero-shot, default inference settings (Qwen3.5 with thinking disabled); higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 37.4 | #266 |
| 2 | Gemini 2.5 Flash | 31.9 | #453 |
| 3 | Gemini 3 Flash | 29.4 | #164 |
| 4 | Gemini 3.1 Pro (Preview) | 26.3 | #94 |
| 5 | GPT-5.1 | 25.7 | #229 |
| 6 | GPT-4o | 17.1 | #741 |
| 7 | Qwen 3.5 9B (Non-reasoning) | 14.7 | #803 (Qwen 3.5 9B) |
| 8 | Qwen 3.5 4B (Non-reasoning) | 12.8 | #1175 (Qwen 3.5 4B) |
| 9 | InternVL3-14B | 7.36 | #1182 |
| 10 | InternVL3-8B | 6.79 | #1688 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=anygroundbench-sports-tvg · How It Works · Data refreshed daily, snapshot 2026-11-13.