Superbash Learn

LLM visual benchmark suite

Same prompts. Different models.

We give every model the same standardized visual brief and compare the results. Pick a model to see what it built.

Anthropic5 runs

Claude Composer 2.5

v2.5

  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Vice City
View runs →
Anthropic5 runs

Claude Fable 5

v5

  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Vice City
View runs →
OpenAI9 runs

GPT Luna 5.6 High

v5.6-high

  • 01 Low Poly World
  • Helms Deep
  • Hogwarts Broom Flight Simulator
  • Jabberwock
  • Office Life
  • Petri Dish
  • Universe Simulator
  • Vice City
  • Yingzao Fashi Assembly
View runs →
OpenAI9 runs

GPT Sol 5.6 High

v5.6-high

  • Helms Deep
  • Hogwarts Broom Flight Simulator
  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Universe Simulator
  • Vice City
  • Yingzao Fashi Assembly
View runs →
OpenAI9 runs

GPT Terra 5.6 High

v5.6

  • Helms Deep
  • Hogwarts Broom Flight Simulator
  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Universe Simulator
  • Vice City
  • Yingzao Fashi Assembly
View runs →
OpenAI6 runs

GPT-5.5

v5.5

  • Hogwarts Broom Flight Simulator
  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Vice City
View runs →
xAI7 runs

Grok 4.5

v4.5

  • Hogwarts Broom Flight Simulator
  • Jabberwock
  • Low Poly World
  • Office Life
  • Petri Dish
  • Universe Simulator
  • Vice City
View runs →