CourtSI-Bench: leaderboard
Metric: Overall score (%): the question-weighted mean of the 13 task scores (threshold mean relative accuracy for distances, binary 30 cm accuracy for localization, exact-match accuracy otherwise) over CourtSI-Bench's 3,686 human-verified spatial questions on single broadcast frames of badminton, tennis and table tennis (players marked by boxes), answers read by exact format match from the model's raw output; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 53.7 | #105 |
| 2 | Seed 1.8 | 52.7 | #136 |
| 3 | Qwen 3 VL 235B A22B | 47.2 | #228 |
| 4 | Qwen 3 Max | 38.2 | #201 |
| 5 | Qwen 2.5 VL 7B Instruct | 37 | #643 |
| 6 | Grok 4 | 36.2 | #169 |
| 7 | InternVL3-8B | 27.8 | #606 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=courtsi-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.