If you run an AI product, the model bill usually arrives as one number. That number hides three separate charges, and each one behaves differently as your product grows.

Cache. A cache hit is input the provider has already seen. On the Standard tier it is $0.10 per 1M tokens. If your product sends the same long system prompt or document on every request, cache pricing is the line item to watch first.

Input. Input is everything the model reads: the prompt, the retrieved context, the file you attach. On the Standard tier it is $0.40 per 1M tokens. It grows with how much context you send, not with how many users you have.

Output. Output is what the model writes. On the Standard tier it is $1.60 per 1M tokens. It grows with the length of the answer, which is why an open-ended chat feature can cost more than a tightly scoped one.

Transcription is a separate meter: it is billed by audio duration, not by text tokens.

A migration checklist

WebWright publishes an OpenAI-compatible API at https://webwright.ai/v1. The documented endpoints are GET /v1/models, POST /v1/chat/completions and POST /v1/audio/transcriptions. The surface follows the OpenAI API shape:

  1. Point your client's base URL at https://webwright.ai/v1.
  2. Send your existing API key in the Authorization: Bearer <your-key> header. WebWright keys use the www_v1_ prefix.
  3. List models with GET /v1/models and pick wren-r1 or wren-r1-fast for chat, or whisper-large-v3-turbo for speech to text.
  4. Send one real request with the payload your product already uses. Buffered JSON and SSE streaming are both documented.
  5. Compare the response shape and the usage block against what your code expects.

If you want a second pair of eyes on the migration, or on which line item is actually driving your bill, you can book a 30 minute call.

The WebWright Team