Pika Labs Releases Open-Weights Video Model Pika H3 Flex, Claiming #1 Among Open Models on DesignArena Leaderboard
Pika Labs has launched Pika H3 Flex, an open-weights video generation model post-trained on MiniMax H3. The model supports multi-shot storytelling, complex camera moves, and audio stem exports. Officially, it ranks #1 among open-weights models on DesignArena’s text-to-video leaderboard and can generate a 15-second 768p clip with synchronized audio in as little as seven seconds.
Pika Labs officially released its new video generation model, Pika H3 Flex, on October 29, 2026. This open-weights model is post-trained on the MiniMax H3 architecture and is designed to give creatives greater speed and control during iteration. According to the official description, the model allows users to bring up to 12 image, video, and audio references, direct complex camera moves, build multi-shot stories, and explore new visual looks.
In terms of performance metrics, Pika Labs claims that Pika H3 Flex can generate a 15-second clip with synchronized audio at 768p resolution in as little as seven seconds. Additionally, the model supports exporting five separate audio stems and finishing videos in up to 4K resolution. The company also states that the model debuted at #1 among open-weights models and #7 overall on DesignArena’s text-to-video leaderboard.
As an open-source project, code and setup instructions for Pika H3 Flex are available on GitHub, while the model weights are hosted on Hugging Face. This release means developers can not only use the model for content creation but also build upon and fine-tune the model itself. Although specific benchmark methodologies and independent reproduction results were not detailed in the announcement, this product release provides a new open-source option and technical reference points for the AI video generation field.