Apple has published three new AI research studies that suggest the company hasn't abandoned spatial computing, despite persistent rumors claiming the Apple Vision Pro is dead. The studies, posted on Apple's Machine Learning blog, explore using large language models for sign language annotation, 3D head modeling, and a new benchmarking system for spatial-functional intelligence.

The most significant study introduces SFI-Bench, a video-based benchmark with 1,555 expert-annotated questions derived from 134 indoor video scans. Unlike existing benchmarks that only test geometric understanding, SFI-Bench evaluates whether AI models comprehend what objects are, where they're located, how they're used, and how they can be repaired. Google's Gemini 3.1 Pro performed best in testing, followed by OpenAI's GPT-5.4-High.

A separate study explores using AI to annotate sign language videos, potentially expanding Apple's Live Translation feature to include sign language. Another develops HeadsUp, a method for creating 3D head models from multi-view captures using an internal dataset of more than 10,000 subjects. This could improve the Vision Pro's Persona feature, which renders the user's face for virtual meetings.