Blender head-to-head: same prompt, 12 seconds, and one frontier model is in a different class
ZeroStateReflex · x · 2026-09-09
- Using the same Blender prompt for 12 seconds each, Claude Fable 5.1's output was clearly subpar compared to GPT-6 Astra — a gap that surprised a developer who builds AI tools inside Blender.
- The author's theory: the model isn't dumb, Blender simply isn't in Anthropic's training distribution yet.
- Analogy to Karpathy's chess observation: early models were terrible at chess until chess data entered the next training run, then improved immensely — the same will likely happen here.
- Takeaway: real in-software tests reveal capability gaps that generic benchmark scores hide. (Models not officially confirmed.)
Related event: Flagship Models Diverge Sharply in Same-Prompt Blender Test(2 posts)→
More from Fun
- $50, 2-hour Kurzgesagt-style AI video reignites debate over intent in AI creation — motionbynick · 2026-10-27
- Fake-paper meme shows an agent workflow for reverse engineering trading systems into a one-person quant fund — PtrPomorski · 2026-10-27
- Steering two orthogonal concept neurons makes a model compose "gnome" — Sauers_ · 2026-10-27
- Fake BTS of a dam disaster made with Dreamina Seedance for 29 cents — Standard_Spell5873 · 2026-10-27
- Meme chart roasts the monogs at San Francisco AI parties — NathanpmYoung · 2026-10-27
- AI video series "Side Quest Demon" Episode IV: The day Mochi escaped — Josepho · 2026-10-27