Models and cost
Style checks, structure checks and Lint Project use no AI and cost nothing. AI costs money only when you use an AI feature: the AI review in Lint Topic, Rewrite, Ask AI, Rewrite… on an issue, Topic Metadata, Add Alt Text, and the AI steps in Import Rules, Rule Decisions and Test Rules.
Which model does what
| Task | Tier | Default model and effort |
|---|---|---|
| AI review, Rewrite, Ask AI | Standard | Claude Sonnet 5.5, medium |
| Alt text, rules from notes | Standard | Claude Sonnet 5.5, medium |
| Rewrite… on an issue, Topic Metadata | Fast | Claude Haiku 5.5, medium |
| Draft a change with AI (owners) | Deep | Claude Opus 5.5, high |
Leave the models on Default unless your style guide owner says otherwise. In tests, Sonnet changed only what each task called for. Haiku sometimes dropped words in rewrites, and Opus added details the text didn't have.
What it costs
An AI review costs a few cents per topic, more for long topics. Saved reviews are reused at no cost. For measured and estimated costs per action, per writer and per team, see docs/testing/ai-cost.md in the House Style repository.
Diagnostics… in House Style > Settings > Advanced shows the models in use and, for this session, the AI calls and estimated cost per task.
Set models for the team
A style guide owner can set models for every writer in the guide's FlareAI/models.json. The team file wins over each writer's Settings.
{
"tiers": { "fast": { "model": "claude-haiku-5-5", "effort": "medium" } },
"tasks": { "review": "standard" },
"review_unit": "section"
}
review_unit:section(the default) reviews a short topic, or one heading's worth of a long one, at a time.blockreviews one paragraph, list item or table row at a time. Sections cost less and run faster.always_packs: the groups of rules every review includes. The default is Headings, UI, Lists and Steps. Other groups, such as the Code rules, are sent only for sections that need them.