Small decision models, built in the UK

Small models.
Sharp instincts.

Very small models that each make one decision well.

They run on a laptop, a phone or the cheapest server you have. No GPU, no cloud round trip. Built in the UK for sovereign use.

Most of what we ask large models to do is sorting, not thinking.

91.8% on 1,400 held-out decisions, ahead of the medium tier of a leading hosted LLM family, at about one 450th of the cost. See the evidence

Myles, a plush dinosaur, looking up at you

The Myles models

One job, done well, on any machine.

In your browser 22M

Runs in your browser or on a phone.

27 ms a decision

Available 150M

Runs on a laptop.

196 ms a decision

Available 400M

Runs on the cheapest server you have.

About 500 ms a decision

In training 1B

Next.

Median time for one decision on one CPU thread of a desktop processor.

All the models and how they are measured

Measured, not claimed

Ahead of the medium tier of a leading hosted LLM family, at one 450th of the cost.

91.8%correct on 1,400 held-out decisions

£4per million decisions on one CPU thread, against £1,880 for the medium hosted tier

74%on multi-rule problems from domains it never trained on, where chance is 16%

59.7%on an independent public benchmark, the highest of the open models under 1B parameters we have tested

Myles-400M on our frozen evaluation set, September 2026. Every model answers the same 1,400 decisions; the hosted models answer zero-shot with one fixed prompt.

Read the method and the full results

Sovereign

Built in the UK. Runs on your hardware. Stays in your estate.

It runs on your hardware. Servers, laptops and devices you already operate, including disconnected networks.
It is built in the UK. Launchables is a UK company, Pitwall AI Ltd. Our models are trained and evaluated on UK-owned hardware.
It learns from UK public text. Legislation, GOV.UK guidance and the parliamentary record, under the Open Government Licence and the Open Parliament Licence.
Its lineage is published. No Chinese-origin models or data anywhere in the lineage, and no outputs of hosted models as training data.
Its errors are measured. A certified error rate on held-out data, with the rest escalated to a person.

For government and regulated teams

Manifesto

The decisions matter more than the chat.

Read the manifesto

Small models.Sharp instincts.

One job, done well, on any machine.

Ahead of the medium tier of a leading hosted LLM family, at one 450th of the cost.

Built in the UK. Runs on your hardware. Stays in your estate.

The decisions matter more than the chat.

Small models.
Sharp instincts.