The Pulse
Higgsfield Says GPT-6 Astra Extends AI Videos Past 10 Minutes
Higgsfield CEO Alex Mashrabov says the company has used OpenAI’s GPT-6 Astra to build workflows that produce instructional videos longer than 10 minutes, up from about one minute. The OpenAI profile also describes Higgsfield’s use of image-
AI.info Team · Written with AI ·
From about a minute to more than 10
Higgsfield says it has moved from AI-generated videos that ran about a minute to instructional videos exceeding 10 minutes. The company attributes the change to GPT-6 Astra, according to an OpenAI customer profile published October 13. Higgsfield CEO and co-founder Alex Mashrabov says the longer videos come from a production workflow built with the model, not simply a new runtime setting.
OpenAI’s account describes Higgsfield’s creative agents handling repetitive work, including generating variations of a shot. With GPT-6 Astra, Mashrabov says, those agents can also call external tools for demanding manual tasks such as 3D modeling. He identifies consistency across a long video as one of the hardest problems for creative AI.
A workflow, not a disclosed benchmark
The example OpenAI gives is Higgsfield’s own instructional-video production. Mashrabov says producing material for the company’s YouTube channel once required considerable manual effort; the team now uses AI to make videos longer than 10 minutes. OpenAI says the channel has nearly half a million subscribers, citing Mashrabov.
The profile does not provide a benchmark, sample video, generation time, cost, or details about how much human editing remains in that workflow. It also does not establish that every Higgsfield user can generate videos of that length. The evidence presented is a company account of an internal production process, rather than an independently tested comparison of video quality or consistency.
Smaller teams are the intended beneficiaries
Higgsfield presents the longer format as part of a broader effort to let small creative teams take on work that once demanded larger production resources. Mashrabov says the company uses GPT-Image-2.5 for promotional images with readable text and infographics, alongside Astra for longer video workflows. OpenAI’s profile frames those tools as useful for small businesses and individual creators, but offers no pricing or access terms for the capabilities described.
“What’s most important for us at Higgsfield is to see that smaller creative teams can actually be more successful now with the power of models like GPT-6 Astra and GPT-Image-2.5,” Mashrabov says. The same account describes Higgsfield’s organization as roughly 200 creatives working with 100 engineers, with Codex helping creatives turn prototypes into production software. The figures and product claims come from the OpenAI profile.
Voice-directed filmmaking is still a stated ambition
Higgsfield’s next step, as Mashrabov describes it, is “voice to movie”: a creator speaks an idea and sees it take shape in real time. He points to OpenAI’s planning, tool-calling, image-generation and voice capabilities as components that could support that experience. The profile gives no launch date or technical specifications for this voice-directed workflow, so it is a stated direction rather than a described product release.
Mashrabov also describes an early educational use for longer videos: he supports rural schools in Kazakhstan and says GPT-6 Astra’s visual understanding could help identify topics students struggle with, then produce tailored instructional material. For now, the concrete claim in OpenAI’s account is narrower: Higgsfield has built a workflow for instructional videos over 10 minutes, where it previously produced videos of about one minute.