December 2025: We published “mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAs”, a new class of Video-Action Model that replaces the VLM backbone of traditional VLAs with a video model, allowing us to leverage pure video pre-training for robot learning.