Astell models and frontier models
What Astell's own chat models are, what frontier models are, per-model token rates with a worked example, what Ultra mode adds, and how the workspace and personal toggles combine.
What are Astell's own models?
Fast and Exacto are Astell's own models, post-trained and orchestrated for how work happens in Astell: following loops, citing sources, calling tools, and remembering context. They are tuned for this product and improve continuously, which is why they carry Astell names. Fast is the standard everyday model; Exacto is the advanced model for work that has to be exactly right. See Astell's AI models for how to choose between them.
What are frontier models?
Frontier models are the leading labs' general-purpose flagships, offered exactly as the labs ship them and always labeled with the lab, the model, and the version. They are for people who specifically want a lab's raw model answering their conversation. When you pick one, it answers your messages; Astell's own models keep doing the supporting work around it (search, research, and other delegated steps).
Frontier models are available on the Grove and Forest plans. The catalog is hand-picked and can change: models are added as labs ship them and retired when they are superseded.
What do frontier models cost?
Every model draws from the same monthly token pool as the rest of Astell, at that model's own rate. Rates are tokens consumed per 1,000 model tokens:
The same rates appear next to each model in the chat model picker, so the cost is never a surprise.
A worked example. Say you send Claude Opus 5 a message that, with your conversation context, comes to 4,000 input tokens, and it returns a 1,000-token answer. Input costs 4 × 3 = 12 tokens and output costs 1 × 15 = 15 tokens, so the turn draws 27 tokens from your pool.
Cached input (context the model has already seen) counts as ordinary input at the model's input rate.
What does Ultra mode add?
Ultra mode is Astell's heavier, more thorough mode: it delegates more work in parallel, runs stronger supporting models, and adds extra checking before the answer is finalized. It works with the Exacto model and with frontier models, and it consumes more tokens because it does more work.
Who can use frontier models?
Two switches combine, and both must be on:
- The workspace toggle, in workspace settings, controlled by workspace owners and admins on Grove and Forest. It is off by default. Turning it on also means that when a member picks a lab's model, workspace content from that conversation is sent to that lab.
- Your personal toggle, in your account settings, on by default. It lets you personally opt out even where your workspace allows frontier models, and it only takes effect in workspaces that enable them.
If either switch is off, the frontier section still appears in your model picker, but locked: it shows a line naming what is holding it, either your plan, the workspace toggle, or your own setting. Nothing in a locked section can be selected, so no usage can happen at frontier rates.
What happens if a model I used is retired?
Nothing breaks. The conversation keeps working; the picker hides the retired model, and your next message runs on your default Astell model with a notice explaining the switch.