SFI-Bench - Layout Inference: leaderboard
Metric: Accuracy (%) on the layout inference questions, four-option multiple choice on egocentric indoor videos (ARKitScenes and ScanNet++ scans), zero-shot with the same prompt templates, models at their default configurations; answered directly with no tools; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 86.8 |
| 2 | Gemini 2.5 Pro | 83.8 |
| 3 | GPT-5.4 | 83 |
| 4 | O4 Mini | 82.4 |
| 5 | GPT-5 | 81.5 |
| 6 | Gemini 3.1 Flash Lite | 81.3 |
| 7 | GPT-5.4 (High) | 81.1 |
| 8 | Qwen 3 VL 235B A22B Instruct | 78.8 |
| 9 | Qwen 3 VL 32B Instruct | 76.7 |
| 10 | Qwen 3 VL 32B (Thinking) | 75.7 |
| 11 | Qwen 3 VL 30B A3B Instruct | 75.5 |
| 12 | Qwen 3 VL 30B A3B (Thinking) | 75 |
| 13 | Qwen 3 VL 235B A22B (Thinking) | 74 |
| 14 | Gemini 2.5 Flash | 73.3 |
| 15 | Qwen 3 VL 8B Instruct | 73.1 |
Interactive version: theaggregate.ai/benchmark?slug=sfi-bench-layout-inference · How It Works · Data refreshed daily, snapshot 2026-10-07.