AI;DR
AI;DR
← Back to live feed1 story

Tuesday, Sep 8, 2026

1

GPT-6 Astra Pro Renders 100.1 Million Blocks in First Model to Handle All MineBench Prompts

OpenAI's GPT-6 Astra Pro produced a programmatic representation of New York City spanning 8.2 kilometers edge to edge. The model operated on an 8,192 cubed grid to generate a 1.65 gigabyte build expanded into JSON. This specific output utilized 100.1 million blocks and originated from a program sized at approximately 61 kilobytes.

The output represents a 32,768 fold increase in possible voxel positions over the 256 cubed grid standard on the MineBench testing platform. MineBench described Astra Pro as the first model in its testing to handle any prompt provided. The project used a scale of one meter per block to create a dense and accurate cityscape.

Sign in to suggest edits

Key sources

  1. SOURCE@minebench_ai“the canvas spans 8.2 km edge-to-edge”x.com
  2. SUPPORT@borismpower“This year I suspect we will see the entire world rendered in 3d!”x.com
  3. SOURCE@nicdunz“first model we've tested that seems to be able to handle basically any prompt”x.com
  4. SUPPORT@nicdunz“solving puzzles that require juggling 5 moving parts with ease”x.com
  5. SUPPORT@scaling01“Astra spent 60+ hours in this 3D open world spatial reasoning eval”x.com
  6. SUPPORT@scaling01“okay that is a disgusting mog if I've ever seen one”x.com
  7. SOURCE@htihle“Astra spent 60+ hours in this 3D open world spatial reasoning eval”x.com
  8. SUPPORT@rohanpaul_ai“reduced a projected 3B-token trajectory to about 350M tokens”x.com
Markdown