Higgsfield: GPT-6 Astra takes AI video past 10 minutes

Higgsfield AI's pitch rests on a specific bottleneck it says GPT-6 Astra removed. Before that model, its videos topped out at around one minute; CEO and co-founder Alex Mashrabov says that ceiling is now over 10 minutes, and the company's own YouTube channel — which he says has almost half a million subscribers — now runs instructional videos past that length.

The company pairs two model upgrades in its story. GPT-6 Astra is credited with longer runs and with letting creative agents call external tools for 3D modeling work, while GPT-Image-2.5 is described by Alex as the first model of its kind that can reliably render text and infographics on top of imagery — the part small businesses need for promotional material. Internally, Higgsfield AI runs about 200 creatives alongside 100 engineers, and Alex says Codex gives the creative side the power of a strong software engineer, letting prototypes move toward production with less engineering involvement.

Looking further out, the company describes a "voice to movie" goal, in which AI understands a creator's spoken intention and renders it in real time. Alex points to OpenAI's work in planning, tool calling, image generation, and voice as what brings that experience closer.