Post-Cutoff.com
  1. Home
  2. Videos
  3. Opus 5.5 vs GPT-6 Astra: one prompt, Blender…

Opus 5.5 vs GPT-6 Astra: one prompt, Blender only, all procedural (X video)

Stefan 3D AI (@Stefan_3D_AI) · 2026-09-22 · ai-made · 469,945 views

▶ Watch on X

Made by AI

Model: Claude Opus 5.5, GPT-6 Astra · Series: Agent-built game or 3D world (the video shows the result)

Evidence: X post (2026-09-22): 'One prompt, Blender only, all procedural... Opus 5.5: 35 min, 199.6k output tokens, about $13.3 in API terms. GPT-6 Astra: 28 min, 56.6k output tokens, about $14.5 in API terms.'

Human role: Same prompt to both; the models built and rendered in Blender and recorded their own build timelapses.

Pipeline: One prompt → each model drives Blender procedurally → 10-second render + build timelapse

What's in the video

Description written by Gemini, which watched and listened to the whole video.

Summary
This video showcases a side-by-side procedural 3D modeling comparison between Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra, created by Stefan 3D AI (@Stefan_3D_AI). Both frontier models were prompted to create a complete, fully procedural Disney-style fairytale castle intro with fireworks, water reflections, and animated text in Blender using only procedural scripting.


What is shown


Claims & numbers


Notable quotes


Assessment

This is an independent technical demonstration evaluating the procedural Blender Python scripting capabilities of Claude Opus 5.5 against GPT-6 Astra. The assets and scenes shown are authentic Blender procedural renders, though presented via edited time-lapses and pre-rendered playback rather than an unedited real-time code-execution feed.


Lyrics & themes


Lore & references


Visual style & craft

Described by gemini-3.8-flash on 2026-10-05 from the video's audio and frames.

Related events