Model updates — LADLE models

AI model providers routinely update their models silently — a company using the API today might get a different-behaving model tomorrow. LADLE opts into the updated versions after a brief testing window, but we log every version change here so subscribers can correlate any behavior shift with a specific model update.

Current model

Heavy tierClaude Sonnet 4.6

Light tierClaude Haiku 4.5

Routing rulePublished at /modelsDeep in the composer forces Sonnet on any turn

ProviderAnthropic, PBC

Training on your chatsNever — prohibited by Anthropic's commercial API termsA zero-data-retention addendum is a goal, not yet signed; a limited abuse-monitoring window applies

Update history

Empty on purpose. LADLE has not launched, so there has been no model change affecting a subscriber — and a version-history table is only worth anything if it was written as the changes happened rather than reconstructed afterwards. The first entry lands here the first time we move either tier to a new version, and the current versions are in the table above.

NO ENTRIES YET · PRE-LAUNCH

How we handle updates

1. Anthropic announces a new version of either tier.
2. We test internally for 3-7 days across a representative set of prompts from our own workflow + a sampled cross-section of user prompts (anonymized, opt-in only).
3. If quality is equal-or-better with no regressions, we roll out to all users on a single day, announce in the newsletter, and log here.
4. If quality has regressions on specific workflows, we hold the previous version longer and file feedback with Anthropic. We do not roll out a version that would materially degrade user experience.

What subscribers see

Nothing changes in the UI when we update the model. Your chats, projects, and instructions all continue to work identically. If you notice a behavior change that correlates with a rollout date on this page, that's real — different Sonnet versions have subtly different voices and reasoning tendencies. Feedback: help@ladle.chat with 'model update feedback' and the date.

What about model deprecation

Anthropic deprecates older Sonnet versions eventually. When they deprecate a version we're still running, we've already moved to the newer one — we don't lag behind their supported models. If you have a long chat that was drafted against an older version and you resume it after an update, you'll get responses from the newer version; the model can't 'switch back' in the middle of a conversation.

SUBSCRIBE TO /NEWSLETTER FOR ROLLOUT ANNOUNCEMENTS

Every model change, dated.