Higgsfield Says Its GPT-6 Astra Workflow Now Produces Videos Over 10 Minutes

OpenAI published the Higgsfield result on October 13

OpenAI published “Higgsfield takes AI video from 1 minute to over 10” on October 13, 2026. The customer story quotes Higgsfield CEO and co-founder Alex Mashrabov saying users can now create videos over 10 minutes long, where the company’s videos previously topped out at around one minute. It says Higgsfield has used that workflow to generate long instructional videos for its own channel.

The announcement does not name a new Higgsfield plan, endpoint, model ID, release cohort, geography, price or service level. Higgsfield’s public homepage does identify its Supercomputer product as an agent powered by GPT-6 Astra, but neither reviewed Higgsfield surface presents an over-10-minute generation control or API contract. The verified status is a vendor-published customer workflow result—not a newly specified public launch.

Astra orchestrates the work; it is not the video renderer

OpenAI says Higgsfield’s creative agents handle repetitive production work such as shot variations and use GPT-6 Astra tool calling for complex manual tasks in 3D modeling. Mashrabov attributes better consistency across longer videos to that workflow. OpenAI’s own model reference, however, documents gpt-6-astra with text and image input and text output. It does not list native video output.

That makes “over 10 minutes” a system-level result. A long finished video can be assembled from planning, scripts, shot generation, image assets, external renderers, editing, retries and human review. Higgsfield’s current public catalog reinforces the distinction: the displayed video endpoints advertise clips up to 15 or 30 seconds. A ten-minute deliverable may be valuable, but it is not equivalent to one continuous ten-minute model generation.

The evidence is a customer account, not a benchmark

OpenAI provides no sample set, before-and-after protocol, project count, success rate, continuity score, rejected-shot count, render time, token use, external-model cost or editor-hours comparison. The story points to Higgsfield’s instructional content as evidence and reports the CEO’s account; it does not publish a reproducible evaluation or independently reviewed production record.

Teams should therefore preserve the narrow claim: Higgsfield says its workflow can produce instructional videos over 10 minutes. Do not convert that into “Astra generates ten-minute video,” a guaranteed duration for every user, a quality comparison with another system or a measured productivity gain. Those stronger statements are not supported by the reviewed material.

Pricing remains split across an undisclosed stack

OpenAI currently lists GPT-6 Astra Standard at $10 per million input tokens, $1 cached input, $12.50 cache writes and $50 output; requests above 272,000 input tokens use higher rates for the full request. Those are Astra token prices only. They do not include Higgsfield subscriptions or credits, video and image generation, 3D tools, storage, rendering, retries or human production time.

Neither the customer story nor the reviewed public Higgsfield catalog gives a price for the described long-video workflow. A buyer cannot infer cost per finished minute from Astra’s token table. The useful procurement unit is total cost per accepted deliverable, with every model call, failed generation, render and editor intervention included.

Image, code and voice claims have different release states

The same story says GPT-Image-2.5 helps Higgsfield create precise campaign visuals and that creatives use Codex to turn ideas into software prototypes. Mashrabov says GPT-6 Astra reduces the engineering work needed to move those prototypes into production. These are customer-reported outcomes; OpenAI provides no defect rate, review burden or comparative study in the article.

The final section is explicitly forward-looking. Higgsfield wants creators to describe an idea by voice and see a movie take shape in real time, and the story connects that ambition to GPT-Live-1 and OpenAI advances in planning and tool calling. It does not announce a voice-to-video SKU, API, preview, access list, date or price. Readers should not report the concept as available now.

What creative teams should do next

Trial the workflow on a bounded format such as a five- to ten-minute instructional video. Freeze the brief and acceptance rubric, then capture the orchestration graph, model and tool versions, generated clip lengths, retries, continuity errors, factual corrections, render latency, token and media charges, editor time and final approval rate. Compare it with the team’s current process on the same deliverable.

Constrain claims to the tested workflow and keep human approval over story, likeness, rights, brand and final cut. Wait for a named public long-video surface and written commercial terms before budgeting around the headline duration. Reject any proposal that treats a stitched workflow as a native video-model capability, omits external-tool costs or labels the future voice-to-video concept as generally available.

Disclosures

AccessAllGPT received no Higgsfield or OpenAI account, credits, subscription, output, project file, benchmark, data, briefing, review, payment or compensation for this article. OpenAI and Higgsfield did not sponsor, review or endorse it. AccessAllGPT did not test, score or rank the workflow. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with OpenAI, Higgsfield or organizations cited. Publication-wide relationships are listed on the disclosures page.