CaST-Bench - Spatiotemporal IoU: leaderboard
Evaluates causal chain-grounded spatio-temporal reasoning for video question answering on CaST-Bench: 2,066 questions over 1,015 videos whose causal chains are annotated as temporal segments with per-second bounding boxes. Spatiotemporal IoU is the mean coverage-aware vIoU* per ground-truth evidence instance — temporal IoU multiplied by the mean per-second spatial box IoU over the overlapping seconds — with unmatched ground-truth instances counted as zero, so spatial imprecision, temporal offset and missed evidence are penalised together. This is the open-submission Space leaderboard for CaST-Bench v1; published fractions are converted to percentages. This series reports CaST-Bench - Spatiotemporal IoU using IoU (%).
Metric: IoU (%). Source: huggingface.co. Saturation forecast: Around July 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 18.73 | #64 |
| 2 | Gemini 3.5 Flash | 15.71 | #66 |
| 3 | Claude Opus 4.7 | 11.32 | #51 |
| 4 | Claude Sonnet 4.6 | 7.93 | #94 |
| 5 | Gemma 4 26B A4B (IT) | 5.52 | #293 |
| 6 | Qwen 3 VL 8B Instruct | 2.55 | #419 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cast-bench-spatiotemporal-iou · How It Works · Data refreshed daily, snapshot 2026-11-05.