SiT-Bench - Global Perception & Mapping: leaderboard

Metric: Accuracy (%; item-weighted over the category's subtasks). Source: arxiv.org. Saturation forecast: Around 2032. 31 models tracked.

Top models

#ModelScore
1Gemini 3 Flash (Preview)35.66
2Qwen 2.5 3B Instruct27.93
3Qwen 3 VL 8B Instruct25.74
4InternVL3-8B22.68
5InternVL3-14B20.19
6Qwen 2.5 VL 7B Instruct19.9
7DeepSeek V3.2 (Non-reasoning)19.68
8Qwen 2.5 VL 72B Instruct19.29
9Qwen 3 4B 2507 Instruct18.95
10Qwen 3 VL 4B Instruct18.81
11Qwen 3 30B A3B 2507 Instruct18.38
12Qwen 3 8B (Non-reasoning)18.2
13GPT-4o17.74
14Qwen 3 VL 32B Instruct15.74
15Llama 3.1 8B Instruct14.28

Interactive version: theaggregate.ai/benchmark?slug=sit-bench-global-perception-mapping · How It Works · Data refreshed daily, snapshot 2026-09-25.