Skip to content
Else Docs

Changing what a tier runs

You can point Fast, Smart or Max at a different model permanently. Here is where, what it affects, and why most people should leave it alone.

View as Markdown

Yes, you can change it — and for most people the honest recommendation is don’t. This page is mostly about why, because the reasons are the same reasons the tiers exist.

Where the control is

Section titled “Where the control is”

Settings → How it works. Under the Engine row there is a collapsed disclosure headed Set an engine per speed tier, with a note beside it saying either All on defaults or how many you have changed. Opening it reveals three rows:

  • Fast engine
  • Smart engine
  • Max engine

Each shows what it currently points at — the word Default, or the model’s name once you have changed it — plus a Change button that opens the full model list. Once a tier is pointed somewhere custom, that row also gains Reset to default, and a Reset all tier engines row appears with a single Reset all button. Nothing here is one-way.

What each tier runs by default, and what that costs, is on the tier’s own page — Fast, Smart, Max — with the full catalogue on Model prices. Auto has no row of its own here, because Auto is not a tier: it chooses among these three.

What changing it actually reaches

Section titled “What changing it actually reaches”

Everything, from then on. This is an account-level setting, not a per-run one, and it is resolved at the moment each run starts. So it applies to:

  • new chats and existing chats alike, including a chat that was already set to that tier
  • work you hand off as a task
  • and Auto. Auto picks among your current tiers, not the factory ones. Point Max at a more expensive model and every run Auto sends to Max gets more expensive too.

Else says this to your face when it applies. Choose a model that costs noticeably more than the tier’s default and the row carries a warning: “This model costs noticeably more than this tier’s default — it also changes what Auto does when it routes here.” Choose one whose tool use has not been measured as reliable, and the row says that instead.

Why you probably don’t need to

Section titled “Why you probably don’t need to”

The tier names are about purpose, not about a particular model. Fast means cheapest acceptable answer; Max means deepest reasoning. Those meanings are what Auto routes against. Repointing a tier does not give you a new capability — it changes which model satisfies a job description you did not write.

Auto already does the thing this looks like it’s for. People usually reach for this control because a specific run wanted a different engine. But Auto varies per request already, and if you want a specific engine for a specific piece of work, the composer’s engine menu is the right size of tool for it — see Switching engines. That choice is scoped to one chat and disappears with it. A tier rebinding is permanent and silent, and it will still be in effect months later when a bill looks odd.

It breaks the pairing you rely on to read prices. The tier pages and Model prices state each tier next to the model it runs by default. Once you have repointed a tier, those pages are no longer describing your account, and nothing in the product reminds you of that when you look at a receipt. You are now maintaining a mapping in your head.

When it genuinely is the right control

Section titled “When it genuinely is the right control”

Two cases, both narrow:

  • You have decided one specific model suits all of your work of a given shape. You have actually compared them and you want it standing, not per chat. Point the tier at it and read the warning it gives you.
  • You want handed-off tasks on a particular model. A task that was started without an engine chosen for it runs on the Smart tier, so repointing Smart is how you change what unattended work runs on across the board.

The one failure mode worth knowing

Section titled “The one failure mode worth knowing”

A model you pointed a tier at can stop being available. When that happens Else does not silently give you something else and say nothing: the tier falls back to its own default for that run, and the reply tells you that it did and that your custom pick wasn’t available. That is the intended behaviour, not an error — but it does mean a rebound tier is one more thing that can quietly revert to acting like the default while you assume otherwise.

If you find yourself checking which model a tier points at before trusting a run, that is the signal to hit Reset all and use the per-chat engine menu instead.

Section titled “Related”