Same 3D coding test across Claude, GPT, Gemini, and Fable

, Fable 5, GPT-5.5, and Gemini were compared under the same conditions. The task was to build a rainy neon cyberpunk alley from scratch in . Each model received the same prompt only once, with no edits.

Every result used the same locked 10-second camera movement. That means the visible differences come from the code each model produced and the scene that code created.

Key points

  • The was rerun after shipped.
  • The task was a rainy neon cyberpunk alley built from scratch.
  • , Fable 5, GPT-5.5, and Gemini were compared.
  • Each model got the same prompt once, with no edits afterward.
  • All outputs used the same 10-second camera movement.

Sources covering this story (2)

Read original