Appearance
Drafting model
Overdesk writes your reply drafts with Claude Sonnet 5.5, a model made by Anthropic. Every call runs on your workspace's own Anthropic API key, so you pay Anthropic directly for what you use and can see that usage in your own Anthropic account.
Which model does what
Each kind of work uses the Claude model suited to it.
| Work | Model |
|---|---|
| Reply drafts and follow-up nudges | Claude Sonnet 5.5 |
| The reviewer's check of each draft | Claude Sonnet 5 |
| Reading what a test site showed on a draft | Claude Sonnet 5.5 |
| Screening out mail that is not support | Claude Sonnet 5.5 |
| Reports and analyses | Claude Opus 5.5 |
| Analyses that keep a list of findings | Claude Opus 5 |
| Triage | Claude Sonnet 5.5 |
| Ask Overdesk answers in Slack | Claude Sonnet 5.5 |
| Issue details drafted from a conversation | Claude Sonnet 5.5 |
Every model thinks before it answers, at its high effort level, whatever that model's own default is. The analyses that keep a list of findings stay on Claude Opus 5 because it found more of the issues in the same conversations when the two were compared.
All of it goes through the same key, the one saved in Settings → AI. Nothing is drafted until that key is set: see Add your Anthropic API key.
When Anthropic's safety checks decline a draft, a report or an analysis, Anthropic can hand it to another Claude model to finish, on the same key. Your Anthropic account may then show usage on a model this table does not list.
Where to see it
The Agent drafts page shows the model beside every draft, in the Model column next to what that draft cost. Open a draft and the model is named again in the line under its title.
For totals across the whole workspace, see Usage.
What shapes a draft besides the model
The model is only part of what makes a draft sound like your team. Your knowledge base decides what a draft knows, and the settings on the Customize AI page decide how it writes: