SFI-Bench - Cross-View Path Reasoning: leaderboard

Cross-view multi-hop path reasoning 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 cross-view multi-hop path reasoning 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: Estimated already saturated. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)83.4#99
2GPT-583#178
3GPT-5.4High82.8#135
4Gemini 2.5 Pro80.7#276
5Gemini 3.1 Flash Lite78.3#371
6O4 Mini73.7#368
7LLaVA-Video-72B70.3#1193
8Gemini 2.5 Flash66.8#463
9Qwen 3 VL 235B A22B Instruct66.6#532
10Qwen 3 VL 32B Instruct64.3#606
11Qwen 3 VL Plus64.3#512
12LLaVA-OneVision-72B64.2#1771
13Qwen 3 VL 32B (Thinking)64#573
14InternVL3.5-14B63.6#1504
15Qwen 3 VL 235B A22B (Thinking)62.4#459

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-cross-view-path-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.