SFI-Bench - Troubleshooting: leaderboard
Metric: Accuracy (%) on the causal hypothesis and troubleshooting 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; models without a search tool, answered offline; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 58.1 |
| 2 | Qwen 3 VL 32B Instruct | 54.6 |
| 3 | Qwen 3 VL 30B A3B (Thinking) | 50 |
| 4 | Qwen 3 VL 8B Instruct | 48.8 |
| 5 | Qwen 3 VL 8B (Thinking) | 47.7 |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 45.3 |
| 7 | Qwen 3 VL 30B A3B Instruct | 45.3 |
| 8 | Qwen 3 VL 32B (Thinking) | 44.2 |
Interactive version: theaggregate.ai/benchmark?slug=sfi-bench-troubleshooting · How It Works · Data refreshed daily, snapshot 2026-10-07.