SiT-Bench - Global Perception & Mapping: leaderboard
SiT-Bench category (Beijing Institute of Technology, 2026): assembling a global mental map from fragmented egocentric descriptions, with scene layout reasoning, panoramic counting across views and cognitive mapping into a grid or JSON map (601 items). Item-weighted accuracy in percent, higher is better; scenes are given only as coordinate-aware text descriptions, never as images.
Metric: Accuracy (%; item-weighted over the category's subtasks). Source: arxiv.org. Saturation forecast: Not forecast. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 35.66 | #141 |
| 2 | Qwen 2.5 3B Instruct | 27.93 | #5206 |
| 3 | Qwen 3 VL 8B Instruct | 25.74 | #982 |
| 4 | InternVL3-8B | 22.68 | #1695 |
| 5 | InternVL3-14B | 20.19 | #1184 |
| 6 | Qwen 2.5 VL 7B Instruct | 19.9 | #1842 |
| 7 | DeepSeek V3.2 (Non-reasoning) | 19.68 | #403 (DeepSeek V3.2) |
| 8 | Qwen 2.5 VL 72B Instruct | 19.29 | #845 |
| 9 | Qwen 3 4B 2507 Instruct | 18.95 | #2174 |
| 10 | Qwen 3 VL 4B Instruct | 18.81 | #1281 |
| 11 | Qwen 3 30B A3B 2507 Instruct | 18.38 | #1078 |
| 12 | Qwen 3 8B (Non-reasoning) | 18.2 | #1838 (Qwen 3 8B) |
| 13 | GPT-4o | 17.74 | #741 |
| 14 | Qwen 3 VL 32B Instruct | 15.74 | #582 |
| 15 | Llama 3.1 8B Instruct | 14.28 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sit-bench-global-perception-mapping · How It Works · Data refreshed daily, snapshot 2026-11-12.