Hosted
The hosted Decision API: endpoint, models, prices, access and what is stored.
The hosted Decision API runs Myles-G 12B and Myles-G 31B for you. Its live status is on the status page.
Endpoint and price
| Model | Price | Batch price |
|---|---|---|
Myles-G 12Bmyles-g-12b | £0.0015 per million input tokens; output tokens free | £0.00075 per million input tokens; output tokens free |
Myles-G 31Bmyles-g-31b | £0.005 per million input tokens; output tokens free | £0.0025 per million input tokens; output tokens free |
Every hosted model is served at https://api.mellix.com/v1/decide with the same keys. Name the model in every request: "model": "myles-g-12b" for Myles-G 12B or "model": "myles-g-31b" for Myles-G 31B. Each also has a pinned id per build, such as myles-g-31b-20261027, for a request that must always reach exactly that build; a retired build answers 410 naming its successor (see Models). Myles-G 31B reads text and images, with a context window of 24,576 tokens: a state too long for it is refused with 422 context_length_exceeded, never trimmed. Myles-G 12B has 32,768 tokens and trims a state too long for its window (see State).
Every hosted model is served from Canada (region CANADA-1); see Data. All prices, what is and is not billed, and worked examples are on Pricing.
The response's usage.input_tokens is what you are billed for. A typical decision is about 700 input tokens, so a million decisions costs about £1.05 on Myles-G 12B. Prices exclude VAT; UK VAT is added at checkout.
The Batch API answers many decisions from one file within 24 hours, at half the live price per input token, from the same credit with the same keys. Use it for work that does not need an answer straight away.
Access
Create a key in the console and buy prepaid credit there, from £5 by card, with optional automatic top-up. Credit you buy never expires; promotional credit may (see Console). Keys work once the account has credit and pause when it runs out, until you top it up.
By default each key can make 1,200 requests a minute, with no daily cap. A workspace can set its own per-key limit in the console. Ask us for a higher limit.
Your data
The endpoint does not store the states, questions or answers of live requests. Its access log records only the method, the path, the status, the time taken and a request id.
A batch has to be kept until you collect it: its lines and results are stored encrypted in Canada, the lines are deleted when the batch finishes, and the results 7 days after you first collect them or 14 days after the batch finished, whichever comes first. See Retention and deletion.
Where each kind of data is kept and for how long, in one table: Data. If your data cannot leave your network at all, run Myles on your own hardware.