SIS-Bench - Object Attribute: leaderboard

Metric: Accuracy (%; 387 object attribute questions; four-option multiple choice on UAV videos (at most 32 frames), zero-shot, open models through vLLM with at most 128 generated tokens and proprietary models through their APIs; higher is better). Source: arxiv.org. Saturation forecast: Around December 2026. 26 models tracked.

Top models

#ModelScore
1Qwen 3 VL 8B Instruct82.2
2Gemini 3 Flash (Preview)80.6
3Qwen 3 VL 4B Instruct80.4
4Qwen 3 VL 30B A3B Instruct78.6
5Kimi K2.578
6InternVL3-14B77.5
7GLM-4.1V-9B (Thinking)77.5
8InternVL3.5-8B76.5
9Qwen 3.5 Plus76.2
10Qwen 3 VL 8B (Thinking)76
11GPT-5.474.9
12Seed 1.872.6
13Step3 VL 10B72.4
14Qwen 2.5 VL 7B Instruct71.1
15Qwen 2 VL 7B Instruct48.3

Interactive version: theaggregate.ai/benchmark?slug=sis-bench-object-attribute · How It Works · Data refreshed daily, snapshot 2026-09-29.