SFI-Bench (Web Search): leaderboard

SFI-Bench (2026) macro-average accuracy over its six tasks for models that may use a web search tool, for example to read a user manual, before answering the operation planning and troubleshooting questions; the other four tasks are answered directly. Percent; higher is better.

Metric: Macro-average accuracy (%) over the six tasks (global and conditional counting, cross-view path reasoning, layout inference, functional association, operation planning, troubleshooting), four-option multiple choice on egocentric indoor videos (ARKitScenes and ScanNet++ scans), zero-shot with the same prompt templates, models at their default configurations; the first four tasks are answered directly and, for the last two, the model may first retrieve external knowledge such as user manuals through its web search tool; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3.1 Pro (Preview)73.8#99
2GPT-5.4High72.1#135
3GPT-569.4#178
4Gemini 3.1 Flash Lite69.4#371
5Gemini 2.5 Pro67.1#276
6O4 Mini66.8#368
7Gemini 2.5 Flash55.3#463

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-web-search · How It Works · Data refreshed daily, snapshot 2026-11-21.