Never hit a usage limit mid-task

  1. 1
    Open ReasonSign in on the web, or in the Mac app.
  2. 2
    Settings › ModelsConnect the plans you already pay for.
  3. 3
    In the composerGive Reason the branch and what's next.

Use Reason on the web, or in the Mac app.

One account runs out. The turn goes on.

In Reason, when the account a turn runs on hits its usage limit, the turn keeps going instead of failing: Reason moves it to the same model on your other connected accounts for that provider, then to the same model on Reason credits if your workspace can pay, and says so in the conversation, for example "hit its usage limit; switched to Reason credits". Claude Code and Codex don't switch on their own; with them, you hand the task over yourself on Reason Network.

Set up fallback for your coding work

  1. Connect the accounts you already pay for

    In Reason, add your model accounts in Settings › Models: subscriptions such as ChatGPT, and API keys. On a computer connected to Reason, its command line also uses the Claude Code and Codex logins already on that computer.

  2. Keep Reason credits as the backstop

    A Reason plan includes weekly usage in credits, and top-ups never expire. When your own accounts are out, the turn continues on Reason credits until they run out.

  3. Work as usual

    When a switch happens, the conversation shows which account ran out and where the turn went, and the next turn starts on your chosen account again.

  4. Bring tasks from Claude Code or Codex

    When Claude Code or Codex runs out on its own, push the branch and tell Reason on the web or in the Mac app what's next. The task continues on your connected accounts, with this fallback behind it.

What Reason does when a limit hits

The rollover is on by default when you use Reason as yourself: in the web and Mac apps, and on the other ways you sign in to Reason. API callers using an API key get it off by default and can turn it on with allow_model_rollover; benchmark sessions never roll over.

Order: the same model on that provider's other connected accounts, then Reason's copy of the same model when the workspace can pay. If the model can't run anywhere, Reason switches to a default model and shows why, instead of leaving the turn waiting.

Only a real usage limit, an expired or refused sign-in, or an exhausted retry budget moves a turn. A short burst of rate limiting retries the same account, up to ten times, so you aren't moved off your plan while it still has usage.

The switch is for that turn. Each new turn starts again on the account you chose, so you're back on your own plan once its limit resets.

What you see

A note in the conversation names both sides, in the form "<account> hit its usage limit; switched to Reason credits", or to your other account, with the model added if it changed. If the account was unavailable rather than out of usage, it says "is unavailable" instead.

If nothing else can run the turn, it stops with "Your <provider> usage limit was reached": the limit is on your own provider account, not your Reason plan, and you can switch model (a Reason-hosted model runs on Reason credits) or wait for the provider's limit to reset.

Reason credits cost $0.04 each and come from your plan's weekly usage or from top-ups.

What it doesn't do

It doesn't change how Claude Code or Codex behave on their own: those tools stop or offer their own options at their limits. To move a task out of them, push the branch and continue it in Reason.

Questions

Does Reason switch from Claude to Codex automatically?

Reason first keeps the same model, on your other accounts and then on Reason credits. It changes model only when the one you chose can't run anywhere, and it tells you when it does.

Do I have to turn fallback on?

No. It is on by default when you use Reason as yourself. API calls with an API key keep the exact model unless they set allow_model_rollover.

Will I end up paying Reason credits for everything?

No. Each new turn starts on the account you chose; Reason credits are used only while your own accounts are out.

Does this work inside Claude Code or Codex?

No. It applies to Reason's own sessions. When Claude Code or Codex hits its limit, carry the task on in Reason's web or Mac app with the branch and your notes.

More guides