What powers each plan
The AI engines behind Free, Plus, and Pro, named plainly — and the principle that decides which one you get when.
Most products won't tell you which AI model you're actually talking to. We'd rather you know. This article names the engines behind each plan and the principle we use to route between them — so you can judge for yourself whether the price is fair.
The principle
The strongest engines go to the artifacts; the efficient ones go to the flow. Everyday conversation happens hundreds of times a week — it needs to be quick, warm, and affordable. The synthesis work — your reviews, your weekly profile diff, the nightly tending of your Living Profile — happens rarely, and it's where depth is actually felt. So that's where the most capable engines are spent.
Plan by plan
- Free runs conversation on Claude Haiku — a fast, genuinely capable model. Free is real Blue with a shorter memory, not a degraded one.
- Plus runs conversation on Claude Sonnet, a step up in nuance and steadiness. Your nightly distillation, Living Profile tending, and weekly reviews run on Sonnet too.
- Pro runs conversation on the same Claude Sonnet — and moves the weekly synthesis up to Claude Opus, the deepest engine we use. Standing intentions execute on fast models, because every action they take waits for your explicit approval anyway; the thinking that matters happened before you tapped.
Two things worth saying plainly:
- What separates Pro from Plus is not a smarter chat partner — it's Blue acting on your behalf, plus deeper synthesis. If someone tells you an upgrade buys "smarter answers" in every message, they're usually selling you something. We route depth where it compounds.
- Engines change as the field moves. When we upgrade one, this page changes with it — the plan promise ("being seen" on Plus, "Blue acts" on Pro) is the constant.
Why we publish this
Blue's memory is inspectable and cited; the engines should be too. A companion asking for your trust shouldn't make you guess what's underneath it. If anything here doesn't match what you experience, tell us — that's a bug, not a quibble.