On the Claude Max plan, switch models by the phase of the work

What does the Max plan actually buy you?

It gives you every Claude model in one place, from the biggest down to Haiku. If you only ever use one mid-size model on it to keep your usage low, you are paying for a range you never touch. Being careful with tokens is a good habit when you pay per token. On a flat plan, it mostly means you get less than you paid for.

Write the plan before you write code

Start with a short product document that describes the architecture: what the parts are, how they talk to each other, and what finished looks like. Every later step works against that document, so whichever model you switch to knows what it is building and why. Without it, every model starts from zero, and switching models just means switching guesses.

Switch by phase

• Ideas and architecture: the strongest model. Mistakes made here are the most expensive to undo.
• Building against the plan: a faster model. The plan does the thinking, and the model does the typing.
• When the faster model keeps getting it wrong: move up a model instead of rewording the same prompt again.
• Review: the strongest model again, on the changes, before anything ships.

There is no reason to use the same model to come up with the idea, design it and write it. The models are not equally good at all three, and you already paid for the ones that are.

When should you use more than one company?

If you want to be tool agnostic, switching between Claude, Codex and others on the same project, that is a fine way to work, and it is cheaper to do it through an account like OpenRouter than through several monthly plans. The Max plan is built for people who want one consolidated place to get the work done. Pick the one that matches how you actually work.

If you build a product on top of these models, the same thinking applies at a bigger scale: do not build your business on one AI company.