A Cobaia Kitchen original series
Authors: Cobaia Kitchen, Claude Sonnet 4.6 Thinking
Everyone tests AI on coding, maths, and language. We test it on banana bread.
The Banana Bread Benchmark is our ongoing experiment in which we pit AI-generated vegan banana bread recipes against our own human-made version — and then actually bake and taste all of them. No leaderboard tables, no abstract scores. Just ovens, overripe bananas, and a group of very willing guinea pigs.
Why banana bread? Because it’s universally beloved, deceptively nuanced, and a surprisingly honest test of whether an AI can produce something genuinely useful in the kitchen. A recipe can be technically complete and still result in something nobody wants to eat. The proof, as they say, is in the tasting.
Each year we select a new set of frontier models, bake their recipes side by side with ours, and report back — on taste, texture, looks, timing, and everything in between.
The Experiments
🍌🍌 Banana Bread Benchmark — 2026
Three new models, a Melodifestivalen final, six taste testers, and the year AI finally caught up. Or did it?
🍌 Banana Bread Benchmark — 2025
The original. Three AI models, one human recipe, and a clear winner. Spoiler: it wasn’t the AI.