CG-Bench - Open-Ended: leaderboard
CG-Bench open-ended question answering on long videos: free-form answers are judged by an LLM aided by the human-annotated clues, reported as accuracy (%); higher is better (Nanjing University and Shanghai AI Lab, 2024).
Metric: Open-Ended Accuracy (%). Source: cg-bench.github.io. Saturation forecast: Estimated already saturated. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 39.2 | #731 |
| 2 | Claude 3.5 Sonnet | 35.6 | #747 |
| 3 | InternVL2.5-78B | 34.2 | #1018 |
| 4 | Qwen 2 VL 72B | 33.7 | #1130 |
| 5 | Gemini 1.5 Pro | 28.7 | #1053 |
| 6 | MiniCPM-V-2.6 | 26.3 | #3095 |
| 7 | GPT-4o Mini | 24.9 | #1627 |
| 8 | Gemini 1.5 Flash | 24.6 | #1671 |
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-open-ended · How It Works · Data refreshed daily, snapshot 2026-11-12.