Pika has launched Pika H3 Flex, a derivative model built on MiniMax H3, releasing the adapter weights and inference code. The model jointly generates audio and video through six-step denoising, accepts multiple input types, and gives creators a tool for shot design and other creative work. In comparative evaluations, it posts the highest scores on multiple audio-video quality metrics among the models tested, ranks eighth on the DesignArena text-to-video leaderboard, and is the top-ranked open-source video model on that list. Since MiniMax H3 was open-sourced in August this year, partners and overseas developers have optimized and released a range of derivative models for different use cases, and have built innovative applications such as Twitch live streaming and AI TV stations. Within one month of its open-source release, H3 topped 24 million downloads, with more than 300 publicly available derivative models—making it one of the most downloaded models worldwide in 2026.