Pika launches H3 Flex, an open-weights video model built on MiniMax H3
Pika unveils H3 Flex, a post-trained version of MiniMax H3 that generates video with audio faster than real time and ships with open weights.
By GennnAi··1 min read
The model cuts generation from 50 steps to six and produces a 15-second 768p clip in seven seconds of transformer computation. It accepts up to 12 references and, through the API, five separate audio stems. Upscaling to 4K remains hosted by Pika. Adapters and inference code are on Hugging Face and GitHub. Flex reaches Pika's creation product on November 2.
Source: Pika, published on October 29, 2026.