Recommended models (your defaults follow Omniscio's picks)

What it is

Which model is best for everyday coding changes often: a new model ships, a cheaper one becomes good enough, a thinking level turns out to cost more than it is worth. Omniscio keeps one recommendation per engine on its own website. This feature keeps your default model and thinking level for each engine in line with it, so you are always starting new sessions on Omniscio's current pick without having to follow the news yourself.

It is on by default, including for people who were already using Omniscio before it shipped.

Where to find it

• Settings → Sessions → Basics → Use Omniscio's recommended models — directly below Default Model.

How it behaves

What gets updated

• Your default model for each engine — what a new session starts on when you do not pick one.
• Your default thinking level where Omniscio recommends one (today: Claude).

Nothing else moves: running sessions, models you pick for a single session, per-project defaults and per-model thinking levels all stay yours and still win over these defaults.

When it updates

About a minute after Omniscio starts, then every 6 hours. Turning the switch on updates at once.

Your own choice always wins

If you change a default model or thinking level yourself — in Settings, with Set as default in a session's model picker, or from the command line — the switch turns off, so the next update never overwrites what you picked. Turn it back on any time to follow the recommendation again.

It never breaks a working setup

• A recommended model this version of Omniscio does not know is skipped.
• A recommended Claude or Codex model your installed Claude Code / Codex is too old to run is skipped until you update it (and retried at the next check).
• If the recommendation cannot be downloaded, or the file is broken, nothing changes.

How it works with the newest-model upgrade

Automatic newest-model upgrade stands aside while this is on — the recommendation decides your default. It still notes each newer model, so turning this off later never springs a stale upgrade on you.

For Omniscio maintainers — changing the recommendation

1. Edit recommended-engine-settings.json (engines.<engine>.model, optional thinkingLevel).
2. Keep each model equal to the app's built-in default for that engine — a test enforces it, so change DEFAULT_MODEL_BY_PROVIDER (or Claude's defaultModel default) in the same change.
3. Deploy Firebase Hosting. Installs pick it up within 6 hours (the file is cached for 5 minutes).

Contract: recommended-engine-settings-sync-contract.md.