Models & usage
Every model and reasoning effort, plus the meters that track context, cost, and your subscription usage.
Choosing a model
The composer's Model pill opens a searchable picker. Atelier offers the full Claude range — Fable 5, Opus 4.8 and 4.7 and 4.6 (each with a 1M-context variant), Sonnet 4.6 (standard and 1M), and Haiku 4.5. Defaults for new chats live in Settings ▸ Claude Code; the pill changes the current session on the spot.
Reasoning effort sets how hard the model thinks: Low, Medium, High (the default), Extra high, and Max — the top two are tinted as a reminder that thinking costs tokens. A Fast mode toggle lives behind the composer's ⚙.
Switching mid-session
Model and effort changes apply to the very next turn — no restart. If you switch to a model whose context window cannot hold the current conversation, Atelier warns you first: Switch anyway, Compact & switch, or cancel.
The context ring
The ring at the composer's right edge shows how much of the model's context window the session has used, with used/total token counts and the session's cost. It turns red inside the configurable buffer zone (Settings ▸ General) — your cue to press its Compact button or run /compact, which summarizes the conversation into a fresh window. Every assistant reply also carries a meta line — model, tokens, cost, and seconds — so you can see what each turn cost as it happens.
Subscription usage
Signed in with a Claude subscription, the sidebar's usage meter tracks your plan's 5-hour and 7-day windows as percentage bars with reset times. Using an API key instead? Billing is per token, so the meter stays hidden — the context ring's cost readout is your guide.