Native to language and numbers
It reads data in its own form, with text alongside to give it meaning.
Build the Other Half
of Intelligence
Language models learned the world of symbols.
Abel is building a Good machine for systems that move, interact and respond to action.
Large language models mastered one half of intelligence: the world of symbols.
The other half is the world that moves. These systems are measured in numbers, carry memory through time and change when someone intervenes.
Their hardest questions have answers that reality can check: What happens next? What causes what? What changes if we act?
The Good machine is built for the questions posed by a dynamic world. It reads language and numbers in their native forms, produces answers that can be verified through calculation and tested against reality, and is co-designed with the hardware it runs on.
Our mission is to build the other half of intelligence.
A Good Machine
Reasoning you can trust
Every answer is a calculation that can be checked.
Software and silicon, co-designed
Its basic unit is an operator, and its chip is designed with it.
Trust between people and machines
The whole system is built to be examined, corrected and relied on.
A Good machine also builds itself. Each one is made by an organization of agents that learns as it goes, and each one leads to the next.
Building it in stages
We begin with expert models: each narrowly specified, independently testable and useful on its own. Together they generate the methods, data and evidence from which a Good Machine can grow.
Simons
Time Series Foundation Models
Reads long histories across related series without turning numbers into text, and returns a full forecast distribution rather than a single guess.
3public forecasting leaderboards near the frontier
Explore moreKov
Causal Discovery Foundation Model
Learns when to trust classical causal methods. It is confident where data identify a direction and stays uncertain where they do not.
≈3Mparameters
Andrey
Exact Causal Discovery for Agents
Runs classical, provable causal discovery methods exactly and in parallel, so an agent can check its reasoning against them.
5,000×+faster in selected method families
Explore moreJoin the team
Abel is a team of mathematicians, researchers and engineers who go unusually deep on hard problems.
If you believe intelligence must go beyond language, come build what comes next.