SIS-Bench - Positional Relationship: leaderboard
Metric: Accuracy (%; 252 positional relationship 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
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 89.7 |
| 2 | Qwen 3.5 Plus | 87.7 |
| 3 | Seed 1.8 | 87.7 |
| 4 | Kimi K2.5 | 87.3 |
| 5 | GPT-5.4 | 86.9 |
| 6 | Qwen 3 VL 30B A3B Instruct | 85.3 |
| 7 | Qwen 3 VL 8B Instruct | 82.9 |
| 8 | Qwen 3 VL 4B Instruct | 82.1 |
| 9 | InternVL3-14B | 81.3 |
| 10 | Qwen 3 VL 8B (Thinking) | 79.8 |
| 11 | GLM-4.1V-9B (Thinking) | 75.4 |
| 12 | InternVL3.5-8B | 73.8 |
| 13 | Qwen 2.5 VL 7B Instruct | 69 |
| 14 | Qwen 2 VL 7B Instruct | 69 |
| 15 | Step3 VL 10B | 67.5 |
Interactive version: theaggregate.ai/benchmark?slug=sis-bench-positional-relationship · How It Works · Data refreshed daily, snapshot 2026-09-29.