Higgsfield says one engineer shipped video features in a day

Higgsfield used GPT-6 Astra to ship new video advertising features in a single day, according to an OpenAI customer story published on 21 September. The claim is Higgsfield's own, reported by the company that sells the model, and it is the kind of number that matters more than most model benchmarks to a small studio: how long a feature takes to exist.

"We are very excited about GPT-6 Astra helping us to deliver new exploration features just within a day," Higgsfield co-founder and chief executive Alex Mashrabov says in the story. "And this now can be done by just one engineer."

Mashrabov attributes the speed to Astra's long-horizon task planning, its ability to plan work across multiple steps, and close collaboration between Higgsfield's creative team and its engineers. Higgsfield, which announced a $400 million Series B at a $5.4 billion valuation on 17 August 2026, sells end-to-end video workflows; the features in question are aimed at small businesses making video advertisements.

The customer-facing example in the story is a prompt: "Take my top-performing ad and generate 100 new variations." With GPT-6 selected, that request produces new creative directions for an existing ad, including versions localised for different countries. Mashrabov's second quote points at the same customers: "Higgsfield also enables smaller businesses to sell more products by generating ads using video AI. And this is where we have seen major improvements with the GPT-6 model."

What is not in the story is as useful as what is. There is no list of the features shipped, no before-and-after timing, no benchmark and no independent test. A vendor case study is evidence of a workflow, not a measurement of it, and the six-figure-versions claim is an example of a prompt rather than a result count.

Higgsfield's own blog documents the workflow in more detail than the customer story does, including the ad-production pipeline it now runs end to end.