SIS-Bench - Spatial Consistency: leaderboard
Metric: Accuracy (%; 195 spatial consistency 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 2028. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.5 | 76.4 |
| 2 | Qwen 3.5 Plus | 75.9 |
| 3 | Qwen 3 VL 4B Instruct | 73.3 |
| 4 | Gemini 3 Flash (Preview) | 71.8 |
| 5 | InternVL3-14B | 67.7 |
| 6 | Qwen 2.5 VL 7B Instruct | 67.7 |
| 7 | Qwen 3 VL 30B A3B Instruct | 67.7 |
| 8 | Seed 1.8 | 66.2 |
| 9 | Qwen 3 VL 8B Instruct | 64.6 |
| 10 | Qwen 3 VL 8B (Thinking) | 64.6 |
| 11 | GPT-5.4 | 64.1 |
| 12 | Step3 VL 10B | 64.1 |
| 13 | GLM-4.1V-9B (Thinking) | 63.6 |
| 14 | Qwen 2 VL 7B Instruct | 63.6 |
| 15 | InternVL3.5-8B | 55.4 |
Interactive version: theaggregate.ai/benchmark?slug=sis-bench-spatial-consistency · How It Works · Data refreshed daily, snapshot 2026-09-29.