Breaking

Higgsfield AI Demonstrates OpenAI's GPT-6 Astra Painting the Mona Lisa Inside Adobe Photoshop

September 5, 2026 at 19:24 EDT

AI Agents
Media Generation
Foundation Models

Higgsfield AI has shown OpenAI's newly launched GPT-6 Astra model recreating Leonardo da Vinci's Mona Lisa inside Adobe Photoshop, directing the software itself rather than simply generating an image from a prompt. The demonstration, which closed with an open invitation for the public to suggest what the model should paint next, is the latest in a run of showcases the creative-AI company has published since Astra's debut this week.

SEPT 2026 · HIGGSFIELD AI

GPT-6 Astra Paints the Mona Lisa Inside Photoshop

Higgsfield AI had OpenAI's newest model recreate Leonardo da Vinci's masterpiece using its own on-screen controls, rather than simply generating an image from a prompt.

72.6%

Desktop-task accuracy on OSWorld 2.0

~47%

Faster than predecessor model Sol

$5.4B

Higgsfield valuation, up from $1.3B

Time to complete a desktop task (OSWorld 2.0)

40 min

GPT-6 Astra

75 min

GPT-5.6 Sol

Average time per task on OSWorld 2.0, a real-desktop-use benchmark.

WHAT DEVELOPERS PRAISE

Direct, autonomous control of real desktop apps
One-shot 3D scene rebuilds from a single photo
Spatial reasoning called a new capability tier

WHERE IT FALLS SHORT

Trails rival Fable 5.1 on coding-quality indexes
Per-task cost up about 75% at max reasoning effort
Early access favored partners over paying users

Safety watch

Astra is the first OpenAI model to hit the Critical cybersecurity tier in the company's own risk framework.

Its new recurrent-depth reasoning style also makes its step-by-step thinking harder for outside researchers to inspect, which is why broad access to its strongest capabilities stays restricted to vetted testers.

Continue reading

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