Unverified claim: 'GPT-6 Astra' hand-draws portrait pixel by pixel in Canva, scores 72.6% on OSWorld 2.0
heypearlai · x · 2026-09-05
An unverified X post claims a model called 'GPT-6 Astra' spent an hour hand-drawing a portrait inside Canva — actually selecting brushes, mixing colors, and placing pixels one at a time rather than generating an image. The post says the same model scored 72.6% on OSWorld 2.0 and set new marks on FrontierMath and ARC-AGI-3. No official confirmation or third-party sourcing exists; treat as a rumor. If true, the takeaway is that computer-use agents can now complete tasks the slow manual way and still nail them.
Related event: Rumored GPT-6 Astra Reportedly Pixel-Paints Portraits in Canva(4 posts)→
More from Models
- The holodeck may end up procedural worlds with a generative lighting and texture pass — dreamwieber · 2026-09-06
- Ethan Mollick: 'Sparks of AGI' paper deserves credit from GPT-4 to GPT-6 — emollick · 2026-09-06
- GPT-6-Astra Shows Striking SVG Generation Skills in Reddit Demo — Ryoiki-Tokuiten · 2026-09-06
- User claims Astra's world modeling in games beats nearly all humans — repligate · 2026-09-06
- Speculation: OpenAI may backtranslate Blender tutorials into computer-use training envs — kalomaze · 2026-09-06
- GPT-6 Astra early verdict: 3D modeling and cross-app computer work are its standout strengths — Scobleizer · 2026-09-06