AI video has long had a length problem. Clips arrive polished and impressive, then end before anyone has said anything. Higgsfield now says it has taken its output from 1 minute to more than 10, which is roughly the difference between a demo reel and something you could call a short film.
What was announced
The announcement, dated 13 October 2026, is short on detail. It names three OpenAI models and gives each a job:
- GPT-6 Astra is tied to longer videos.
- GPT-Image-2.5 is tied to precise visuals.
- GPT-Live is described as a path to voice filmmaking.
The headline number is the stretch from 1 minute to over 10. That is the only figure on offer.
What we can't tell you yet
The summary is all that was available, so the interesting questions go unanswered. Nothing in it covers how Higgsfield keeps characters and scenes consistent over ten minutes, what the videos cost to generate, or how long they take to render. There are no sample clips, no benchmarks, and no word on pricing or availability.
It also doesn't say whether the 10 minutes is a hard ceiling or just the point where the announcement stopped. "Over 10" leaves room in both directions.
Treat the claim as a claim for now. A minute of video can hide a lot of sins, and ten minutes is much less forgiving.
Why the three-model combination matters
The framing is a division of labour. One model handles duration, another handles how individual frames look, and a third points toward making films by speaking to them. The summary does not explain how the pieces connect inside Higgsfield's product, so any description of the pipeline would be guesswork.
The voice element is worth watching, though. The phrase "a path to" suggests this is a direction of travel, not a finished feature. Whether it is ready for working creators is not something the source says.
The takeaway
If the 10-minute figure holds up in real use, Higgsfield will have moved AI video from a novelty format to something with room for a beginning, a middle and an end. That is a bigger leap than it sounds, mostly because the middle is where AI video has tended to fall apart. We'll know more when someone shows us the full ten minutes.