Friday30 Oct 2026
Pika open-sources six-step Pika H3 Flex video-and-audio model
Released on 29 October 2026, Pika H3 Flex post-trains MiniMax H3 down from 50 denoising steps to six, generating 15-second 768p stereo-audio video clips in seven seconds of transformer compute alongside open weights on Hugging Face.
Open sourceStory 5 of the day
