He Kaiming's Team Releases VISTA, a Visual-Native Harness for Multimodal Models

He Kaiming's team proposed VISTA, a visual-native harness that lets existing multimodal models observe environments, store raw frames, and revisit, zoom in on, and inspect details at inference time without retraining. With VISTA, Claude Opus 5.0 completed all 25 public games in ARC-AGI-3 with a perfect 100 on relative human action efficiency, using 57.4% fewer game actions than the human first-play baseline, while GPT-5.6 Sol also cleared all games with a score of 99. The team further tested the method on GameWorld, AI GameStore, and BabyVision, and listed embodied tasks as future research.