Pricing
There's no subscription. Each message is paid from your balance, and you can see what it costs before and after you send it.
Before you send
The model picker shows each model's live rate in dollars per million tokens (input · output). The rate already includes the protocol fee.
Some operators discount cached input: prompt tokens the model can serve from a cache, which is common when you keep chatting in one conversation. For those models, the pricing detail adds a Cached input line with the lower rate, so you can see the discount before you send; some operators make cached reads free. Models without a cache discount don't show the line.
A few models charge more for long prompts. Their pricing detail adds an Above N line. Once a prompt reaches that many tokens, the whole request (input, cached input, and output) bills at that line's higher rates, not just the part past the threshold. The cached-input rate steps up too, shown in a second Cached input row under the tier. The tier is fixed from your prompt size when you send and can't be raised afterwards. An operator that finds your prompt never reached the threshold bills the base rate.
When you send, a maximum for that message is briefly held from your balance — never more than the rate you were quoted — and whatever the reply doesn't use comes back when it settles.
After each message
Every assistant reply shows what it actually cost in its footer (for a short chat turn, typically a fraction of a cent). Hover the amount to see the breakdown: the operator's base charge plus the protocol fee, and the number of tokens billed. If the model served part of your prompt from its cache, the hover also shows how many tokens were cached (discounted), billed at the operator's lower cache rate.
Charges are backed by operator-signed receipts and settled on-chain against your balance. Before each request your client gets a signed price ceiling, and it rejects any receipt that exceeds it — see how a request is paid.
Free models
Some operators serve a model at no charge: it shows a rate of $0, and the reply's footer bills nothing. Free messages are limited to 20 per day per account, across all operators. The day starts at your first free message, not at midnight: send one at 3pm and your allowance refreshes at 3pm tomorrow.
If you run out, a message tells you so and when your free messages come back. Paid models are unaffected, so you can keep chatting on those at any time.
Free messages aren't billed, but they need your account set up on-chain. A small, refundable deposit backs each request while it's in flight and comes back when it settles. Your first free message also sets aside a one-off fraction of a cent in ALGO that isn't returned; it's what ties your daily allowance to your account. The app handles both for you when you fund. See Funding.
Image generation
Image generation is priced per image. When a request allows image output, the cost of your Images per request limit (see Settings) is reserved up front and anything unused is refunded: a request that allows up to 4 images but produces 1 pays for 1.
Where the money lives
Your balance is dollars (USDC) in a wallet only you control. See Funding and Why use crypto?.