CG-Bench - Accuracy@IoU: leaderboard
CG-Bench credibility metric: a multiple-choice answer counts only if it is correct and the model's cited time interval overlaps the annotated clue above an IoU threshold, averaged over thresholds 0.1 to 0.5 (%); higher is better (Nanjing University and Shanghai AI Lab, 2024).
Metric: Accuracy@IoU (%). Source: cg-bench.github.io. Saturation forecast: Around 2029. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 4.33 | #731 |
| 2 | Qwen 2 VL 72B | 3.17 | #1130 |
| 3 | Claude 3.5 Sonnet | 2.73 | #747 |
| 4 | Gemini 1.5 Pro | 2.64 | #1053 |
| 5 | Gemini 1.5 Flash | 2.49 | #1671 |
| 6 | InternVL2.5-78B | 2.46 | #1018 |
| 7 | GPT-4o Mini | 2.13 | #1627 |
| 8 | MiniCPM-V-2.6 | 0.94 | #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-accuracy-iou · How It Works · Data refreshed daily, snapshot 2026-11-12.