Models
Four models, one router, the cheapest that can do the job
You never pick a model by hand. The Builder classifies each turn and routes it: a question goes to the fast model, a page build to the workhorse, a hard design problem to the deep one. You see which model each turn used and what it cost, per site, in your account.
Haiku
Fastclaude-haiku-4-5-20251001
Questions, small edits, routing decisions.
The Builder's default for anything short: "make the hero shorter", "what does this setting do", deciding which tool to call next. Answers in a second and costs a fraction of a cent per turn.
- ✓Chat questions and explanations
- ✓Small content edits
- ✓Choosing tools and routing
- ✓Menus, metadata, alt texts
input
$1.20
output
$6.00
cached
$0.12
USD per million tokens, platform fee included, plus $0.01 per call
Sonnet
Workhorseclaude-sonnet-5
Building pages, writing copy, shop setup.
The model that builds your first draft: page structure from the brief, copy in the client's language, WooCommerce setup, the design system in theme.json. Strong, fast enough, and a first draft still costs well under a dollar.
- ✓Brief to site
- ✓Page and section copy
- ✓Design system and patterns
- ✓WooCommerce categories and products
input
$3.60
output
$18.00
cached
$0.36
USD per million tokens, platform fee included, plus $0.01 per call
Opus
Deepclaude-opus-5
The hardest design and structure work, on request.
When a brief is complex, a design must be reasoned through, or a whole information architecture has to be planned, the router escalates here. Used sparingly, because it is the expensive one.
- ✓Complex briefs and site architecture
- ✓Design variants to compare
- ✓Long-form pages
- ✓Difficult migrations of content
input
$18.00
output
$90.00
cached
$1.80
USD per million tokens, platform fee included, plus $0.01 per call
How the router decides
A small classifier trained on anonymised builds labels each turn: question, edit, build, design. Each label maps to a model. Confidence below a threshold falls back to the workhorse, never to the expensive one.
Why cached input matters
The site profile, the brand kit and the conversation so far are sent on every turn. The agents mark them for caching, so on the second turn they cost a tenth. Long jobs get cheaper as they go.
Your own agent, your own model
Agents built on the platform for your own product choose from the same four models, with the per-token prices above and the same statement per tenant. Ask us.