Pay for tokens, not for access.
There is no subscription. You either spend Studio credit, which costs the provider’s price plus 25%, or you connect your own key and Studio takes nothing at all.
Free
$0.50 credit
On sign-up. No card.
- 30 messages an hour
- Models up to $15/M output tokens
- $1 a day ceiling
- Knowledge base, personas, branching
- Not included: Image generation
- Not included: Code execution
- Not included: Studio Code projects
Pay as you go
Tokens + 25%
Top up any amount. Credit does not expire.
- Every model in the catalogue
- 300 messages an hour
- Image generation and code execution
- 5 Studio Code projects, 120 minutes of commands a day
- $50 a day ceiling by default, adjustable
Your own key
$0 to Studio
Your provider bills you directly.
- OpenRouter, or a vendor key per provider
- No markup, no model ceiling, no daily cap
- Encrypted at rest, never returned to the browser
- Mix freely: per-chat, not per-account
- Not included: Code execution and Studio Code, which cost server CPU rather than tokens
What that works out to
Model prices move, so these are illustrations rather than a rate card — the live price per million tokens is next to every model in the picker, and the exact cost of every message is on the message.
An hour of back-and-forth
about $0.40
On a mid-tier model, thinking something through with a persona and a few documents in play.
A long document, summarised
about $0.05
Forty pages in, two pages out, on a fast cheap model.
An afternoon in Studio Code
about $2–5
Planning, editing, running tests. Most of it is the coder model re-reading files.
Where the 25% goes, and when it does not apply
When you spend Studio credit, Studio pays the provider and charges you their price plus 25%. That margin is the entire business model — it pays for the server, the sandbox and the person answering your email. There is no second charge anywhere.
When you use your own key, that turn never touches the ledger. Studio makes nothing on it. The daily cap and the model ceiling do not apply either, because you are spending your own money at your own provider.
Code execution and Studio Code are the exception, and it is a fair one: they burn server CPU rather than model tokens, so they need a paid account even if every model call is on your key.
Questions people ask first
- Do I need an API key to start?
- No. Signing in gives you $0.50 of credit, which is enough for a real conversation on a good model. Connect a key whenever you want to.
- What happens when my credit runs out?
- Sending stops with a message telling you so, rather than failing mid-answer. Nothing is deleted and nothing auto-charges. Top up, or connect your own key, and carry on in the same chat.
- How is this different from OpenRouter's own chat?
- OpenRouter routes the request; Studio is the workspace around it. Saved personas, goals, an indexed knowledge base, branching, a Python sandbox, document generation, and a coding agent with its own container. And it can bill to your OpenRouter key, so you are not choosing between them.
- Can I use my ChatGPT or Claude subscription?
- No, and nothing can. Those subscriptions cover the vendor’s own apps and do not come with API access. An API key from the same provider is a separate, usage-billed thing — that is what Studio can use.
- Does credit expire?
- No. It sits on your balance until you spend it.
- Can I get a refund?
- Unspent credit, yes. See the refunds page for how and how quickly.