SFI-Bench - Layout Inference: leaderboard

Room layout inference questions from SFI-Bench (2026): four-option questions about egocentric videos of scanned indoor rooms, answered zero-shot from the video. Accuracy in percent; higher is better.

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 February 2027. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)86.8#99
2Gemini 2.5 Pro83.8#276
3GPT-5.483#135
4O4 Mini82.4#368
5GPT-581.5#178
6Gemini 3.1 Flash Lite81.3#371
7Qwen 3 VL 235B A22B Instruct78.8#532
8Qwen 3 VL 32B Instruct76.7#606
9Qwen 3 VL 32B (Thinking)75.7#573
10Qwen 3 VL 30B A3B Instruct75.5#895
11LLaVA-Video-72B75.2#1193
12Qwen 3 VL 30B A3B (Thinking)75#719
13Qwen 3 VL 235B A22B (Thinking)74#459
14InternVL3.5-30B-A3B74#1267
15Qwen 3 VL Plus73.6#512

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=sfi-bench-layout-inference · How It Works · Data refreshed daily, snapshot 2026-11-21.