Models
The six Myles models, from a browser to one GPU, and how to choose between them.
One family, one job: the decision. Every Myles model reads a state, a question and options and returns a probability for every option. None writes text. Myles-G is called over HTTP, and the CPU models run in your own program. Pick the size for the hardware you have, then see Model types for how to call it. For a side-by-side comparison with a chooser, see the Models page.
Complete library
| Model | Runs on | Median decision | Status |
|---|---|---|---|
| Myles-G 31B | One data-centre GPU | 48 ms | Available |
| Myles-G 12B | One workstation GPU | 31 ms | Available |
| Myles-G Edge | Laptop- and phone-class GPUs | 21 ms | By request |
| Myles-400M | The cheapest server you have | About 500 ms | Available |
| Myles-150M | A laptop | 196 ms | Available |
| Myles-22M | A browser or a phone | 27 ms | Available |
Median time for one decision: Myles-G on one GPU, the others on one CPU thread of a desktop processor.
How you call each model, and where it is available:
- Myles-G 31B and Myles-G 12B:
POST /v1/decide, on the hosted API or on your hardware. - Myles-G Edge:
POST /v1/decide, on your hardware, by request. - Myles-400M and Myles-150M:
decide()in Python, on your hardware, by request in the console. - Myles-22M:
decide()in Python, on your hardware, by request in the console, and in the browser.
"Your hardware" is under a licence per deployment. See Enterprise or talk to us about self-hosting.
Myles-G
Myles-G is for the hardest decisions. Mellix trains it for decisions only, and it runs on one GPU: on a GPU you own, no customer data leaves your premises, and on the hosted API requests are processed in Canada. Myles-G 12B and Myles-G 31B are available on the hosted API, and every Myles-G model is available by request on your own hardware. Every Myles-G model is called the same way: POST /v1/decide with a state and typed questions (see the API reference). Myles-G 12B and Myles-G 31B also read images sent with the request (see Images).
Myles-G is adapted from an openly licensed base model (Apache 2.0), which its lineage record names.
CPU models
Myles-400M, 150M and 22M need no GPU and run on disconnected networks. Myles-400M is 91.8% correct on our frozen set of 1,400 held-out decisions, on one CPU thread (measured on the September 2026 Myles-400M build).
Organisations request access to all three in the console, where they manage API keys. We review every request, and approved organisations download the model files from the console (see Model access). Production use is under a licence per deployment, so talk to us about self-hosting. Myles-22M is small enough to run in a browser.
The CPU models have no HTTP API. You load the model file in your own program and call decide(state, question, options), which returns a probability per option (see the example).
Lineage
Every new Myles model ships with a record of its base model, data and licences. Our standard is no Chinese-origin models or data anywhere in the lineage, and no outputs of hosted models as training data.