CG-Bench - Clue MCQ: leaderboard
CG-Bench multiple-choice accuracy (%) when the model sees only the annotated clue interval of the long video rather than the whole video; higher is better (Nanjing University and Shanghai AI Lab, 2024).
Metric: Clue-Interval MCQ Accuracy (%). Source: cg-bench.github.io. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL2.5-78B | 59.5 | #1018 |
| 2 | GPT-4o (2024-08-06) | 58.6 | #731 |
| 3 | Qwen 2 VL 72B | 57.8 | #1130 |
| 4 | Claude 3.5 Sonnet | 56.5 | #747 |
| 5 | Gemini 1.5 Pro | 50.9 | #1053 |
| 6 | GPT-4o Mini | 48.4 | #1627 |
| 7 | Gemini 1.5 Flash | 48.4 | #1671 |
| 8 | MiniCPM-V-2.6 | 44.4 | #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-mcq · How It Works · Data refreshed daily, snapshot 2026-11-12.