CG-Bench - Clue Grounding mIoU: leaderboard
CG-Bench clue grounding: the mean temporal IoU (%) between the time interval a model cites as evidence for its answer and the annotated clue interval in a long video; higher is better (Nanjing University and Shanghai AI Lab, 2024).
Metric: Temporal Grounding mIoU (%). Source: cg-bench.github.io. Saturation forecast: Around 2029. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 5.73 | #731 |
| 2 | Claude 3.5 Sonnet | 4.17 | #747 |
| 3 | InternVL2.5-78B | 3.9 | #1018 |
| 4 | Gemini 1.5 Pro | 3.85 | #1053 |
| 5 | GPT-4o Mini | 3.68 | #1627 |
| 6 | Qwen 2 VL 72B | 3.64 | #1130 |
| 7 | Gemini 1.5 Flash | 3.43 | #1671 |
| 8 | MiniCPM-V-2.6 | 2.27 | #3095 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cg-bench-clue-grounding-miou · How It Works · Data refreshed daily, snapshot 2026-11-12.