Breaking

Higgsfield Uses GPT-6 Astra to Build an Interactive 3D Engine That Disassembles on a Slider

September 7, 2026 at 12:08 EDT

Foundation Models
Media Generation
AI Agents

Higgsfield used OpenAI’s GPT-6 Astra to generate a detailed, interactive 3D engine that can be taken apart with a slider, turning a mechanical assembly into a part-by-part inspection rather than a static render.

Sep 7, 2026 · Higgsfield × OpenAI

An engine you can take apart

Higgsfield asked GPT-6 Astra to generate a detailed interactive 3D engine. A slider peels the assembly into inspectable components instead of leaving a static render.

How the stack produces the demo

Prompt

Ask for a detailed mechanical 3D model.

→

GPT-6 Astra

Reasons about parts, layout, and motion.

→

Higgsfield MCP

Generates assets and hosts the 3D scene.

→

Slider 3D

Drag to disassemble, then put it back.

3D Jutsu message cost

Credits per agent message. Column height is one block per credit.

GPT-5.6 Sol

~11

Claude Fable 5.1

~17

GPT-6 Astra

Auto, the default, is free. Picking GPT-6 Astra is the highest listed click in the 3D Jutsu selector.

1,025

Individually modeled objects in a related Higgsfield camera exploded-view demo

20%

Fewer tokens vs other models Higgsfield tested, per CEO Alex Mashrabov

95.9%

BenchCAD geometric overlap for GPT-6 Astra with tools, vs 83.3% for Sol

What is landing

Builders call Astra unusually strong on spatial and mechanical work. A slider that dissects an engine is being treated as a teaching and product-demo tool, not only a visual stunt.

What is holding it back

Astra is the costliest listed 3D Jutsu model at about 17 credits per message. Testers also report a repetitive visual style and writing rated below GPT-5.6 Sol, with API list prices about 2.5× Sol.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.