K9-Bench - Open-Ended Overall Quality: leaderboard
K9-Bench also scores free-form answers to its canine video questions with a GPT-4o judge, which rates overall response quality on a zero to ten scale that is rescaled to 0-100. The judge sees the question, the reference answer and the model answer, and also rates logic, factuality, conciseness and accuracy separately.
Metric: Overall response quality (0-100): GPT-4o judge rating, on a 0-10 scale rescaled to 0-100, of the overall quality of each free-form answer to the K9-Bench video questions; 32 uniformly sampled frames for GPT-4o and the open models, the full video at 1 FPS for Gemini 2.5 Pro and Qwen3-VL-235B-A22B. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 69.15 | #266 |
| 2 | Qwen 3 VL 235B A22B Instruct | 67.96 | #522 |
| 3 | Qwen 3 VL 8B (Thinking) | 63.27 | #887 |
| 4 | GPT-4o (2024-08-06) | 60.99 | #740 |
| 5 | Qwen 3 VL 32B (Thinking) | 60.18 | #559 |
| 6 | Qwen 3 VL 32B Instruct | 55.73 | #585 |
| 7 | Qwen 3 VL 8B Instruct | 52.64 | #981 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=k9-bench-open-ended-overall-quality · How It Works · Data refreshed daily, snapshot 2026-11-13.