Higgsfield says OpenAI’s GPT-6 Astra supports AI-generated instructional videos exceeding 10 minutes, alongside image and prototyping tools for creative teams.
OpenAI reported on October 13, 2026, that video platform Higgsfield AI is using GPT-6 Astra in a workflow that produces instructional videos longer than 10 minutes. Higgsfield CEO and co-founder Alex Mashrabov said videos made before the model typically topped out at about one minute.
The workflow uses Higgsfield’s creative agents for repetitive production tasks, including generating alternate versions of a shot. OpenAI says GPT-6 Astra agents can also use external tools for complex manual work such as 3D modeling. Mashrabov linked these capabilities to the challenge of keeping visual details consistent across longer videos.
Higgsfield is also using GPT-Image-2.5 for promotional material. Mashrabov said the model can render text and infographics in images, which he sees as useful for small businesses without large production budgets. The company’s product development involves around 200 creatives and 100 engineers; OpenAI says creatives use Codex to turn concepts into software prototypes, with GPT-6 Astra reducing the work needed to move those ideas toward production.
The examples describe a mix of video generation, image creation and software development tools in one creative workflow, rather than a standalone video-model launch. OpenAI’s account also describes an early education-related use: Mashrabov supports rural schools in Kazakhstan and said GPT-6 Astra’s visual understanding could help identify learning topics students struggle with and generate tailored videos.
Higgsfield’s next goal is a voice-led filmmaking experience in which creators describe an idea aloud and see it take shape. Mashrabov called this vision “voice to movie”; the article presents it as an ambition, not a currently available feature. OpenAI identifies GPT-Live-1 as part of that direction but gives no release timeline.