The latest collaboration between OpenAI and the AI video generation platform Higgsfield has pushed a long-standing boundary significantly further—raising the upper limit of AI-generated video duration from 1 minute to 10 minutes. The key to this leap is Higgsfield's integration of GPT-6Astra. Its most critical capability is enabling "creative agents" to call on external tools to handle complex tasks, effectively providing a video generation process with a self-organized assistant director, ensuring consistency in character appearance, scenes, and visual details throughout a long video, thus avoiding the "casting mix-up" where characters' faces change abruptly.

Higgsfield has already applied this process in practice. Its YouTube channel, which has nearly 500,000 subscribers, has started producing teaching videos longer than 10 minutes. Previously, the work required manual assembly and proofreading, but the model now handles it all at once, saving many resources. At the same time, Higgsfield has also integrated GPT-Image-2.5, specifically responsible for rendering readable text and infographics reliably—this is especially beneficial for small businesses, as it means they no longer need to go to great lengths for promotional materials, and can produce quality videos at low cost.