K9-Bench - Open-Ended Overall Quality: leaderboard

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 December 2026. 9 models tracked.

Top models

#ModelScore
1Gemini 2.5 Pro69.15
2Qwen 3 VL 235B A22B Instruct67.96
3Qwen 3 VL 8B (Thinking)63.27
4InternVL3.5-8B62.85
5GPT-4o (2024-08-06)60.99
6Qwen 3 VL 32B (Thinking)60.18
7Qwen 3 VL 32B Instruct55.73
8Qwen 3 VL 8B Instruct52.64

Interactive version: theaggregate.ai/benchmark?slug=k9-bench-open-ended-overall-quality · How It Works · Data refreshed daily, snapshot 2026-09-29.