Skip to content
VibeCodeBench

VS ·The Flan Test

Opus 5.5 vs Sonnet 5.5 vs Fable 5.1Names hidden. Which is which?

Three Claude models built the same wobbly caramel flan. Poke each one and compare.

Is it fair?

A fair match: same effort, same kind of tool.

  • Same effort
  • Agents only

All three ran at high effort as Claude Code subagents, with the same prompt pasted as is. A workflow restart ran this prompt twice, so each run has 2 tries on record; the first try worked every time and is the one shown.

What these mean
Same effort
Every run used the same step on the shared effort ladder.
Agents only
All runs used coding agents.

Each run

  • AnthropicOpus 5.5Effort: High, level 3 of 5works · try 1 of 2Receipt and tries
  • AnthropicSonnet 5.5Effort: High, level 3 of 5works · try 1 of 2Receipt and tries
  • AnthropicFable 5.1Effort: High, level 3 of 5works · try 1 of 2Receipt and tries

The benchmark

Poke, stretch and scoop a 3D caramel flan with real soft-body physics.

Try this: Tap Shake the plate, then scoop it with the Spoon.

I keep the exact prompts private for now. See it in the gallery.

Make your own comparison

More comparisons