Update

Introducing Treepai

3 October 2026

Treepai is a new AI model provider and lab building specialized language models: models meant to each do a specific task very well, and to cost less to train and to run.

Why cost matters

Training and running modern AI takes enormous amounts of computing power. That cost shapes who gets to build AI, who can afford to use it, and how much energy it uses. We think efficiency deserves to be a research goal in its own right, not something fixed after the fact.

What we’re building

Our goal is models that run where they’re needed, on a company’s own servers (even ordinary CPUs) or through a hosted API. We train from scratch on documented data with clear provenance, including AI-generated examples that carry an open license and whose origin we can verify. Our first research report is now published. Nothing is available to use yet.

How we’ll share our work

We won’t make claims about our models until we can back them up, and when we publish results we’ll explain how we measured them. This page is where we’ll post news as it happens.

To get in touch, email hello@treepai.com.

Updated 3 October 2026 to describe our focus on specialized models, 6 October 2026 to describe our training data, and 7 October 2026 to link our first research report and to describe Treepai as a lab and model provider.

← All updates

Research updates

We email when we publish something new. Unsubscribe at any time. Privacy notice.