SIS-Bench - Relative Direction: leaderboard

Metric: Accuracy (%; 200 relative direction 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 June 2027. 26 models tracked.

Top models

#ModelScore
1Kimi K2.580.5
2Qwen 3 VL 30B A3B Instruct79
3Qwen 3.5 Plus78.5
4Seed 1.876.5
5Gemini 3 Flash (Preview)75
6Qwen 3 VL 8B Instruct73.5
7Qwen 3 VL 8B (Thinking)72.5
8GLM-4.1V-9B (Thinking)72
9InternVL3-14B71
10Qwen 3 VL 4B Instruct71
11InternVL3.5-8B70.5
12Qwen 2.5 VL 7B Instruct69
13Step3 VL 10B68.5
14GPT-5.467
15Qwen 2 VL 7B Instruct61

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