Being-H0.7: A Latent World-Action Model from Egocentric Videos
Original research
Uses future-informed latent reasoning without generating future video frames at inference.
Being-H0.7 trains a latent reasoning interface using current and future observations, then discards the future-observation branch at inference. Its authors explicitly call it a latent World-Action Model.
Core research identity and broad method were source-checked as of 2026-11-03. Independent benchmarking, commercial availability, weights and licensing have not been assessed.