Models & pricing

Every model is served from the same base URL, so switching model is a string change.

• GET https://api.straitly.ai/v1/models lists exactly what your key can call right now. This is the list to trust.
• The Models page is broader: it also carries the per-token price of models we have not opened yet, and marks which ones are callable today.

The four rates

Billing is prepaid and per token. Every request is charged on four separate meters:

• Input: fresh input tokens.
• Cache read: input served out of the cache.
• Cache write: input written into the cache.
• Output: everything the model generates.

Cache writes are the one people forget

• On a prompt-caching workload they can be a double-digit share of the bill, so the console names them separately everywhere: their own Activity column, their own Overview line, their own CSV columns.
• Both TTLs (5 minute and 1 hour) bill at the same cache-write rate. The console reports which TTL you used, so you can see how your cache is behaving.

What you are not charged for

• Failed requests are not billed.
• There is no subscription and no minimum.