Est. 2026 · Los Angeles, CA

Developing models that refine themselves.

Arch advances how AI reasons. We develop systems that think in the open, verify their own work, and refine themselves, toward something that understands what it knows and repairs what it gets wrong.

Get in touch
percentof requirements the model had never seen, met and checked by physics simulation.
percentless computation at equal accuracy, measured against a fixed baseline.
longerproblems than it trained on, solved by thinking longer at run time.

Why we exist

Reasoning you can trust is reasoning you can check.

01

Reason in the open

Reasoning that can be viewed, not a sealed answer.

02

Verify its work

Knowing when it is right, and when it is not.

03

Refine itself

Sharper with every iteration, and measured at each one.

From the research

Proven, written to be read.

View all →
July 2026

A small AI that solves problems it was never trained on, by thinking longer.

Held to the amount of thinking it used in training, it got 11 of every 100 right. Allowed to keep thinking, it got essentially all of them. The full report, with every baseline and what is not proven yet.

Read the full report →

Contact

The work is early and mostly private.

If you want to know what we are developing or where it stands, from a simple question to a serious conversation, write to us.

[email protected]