The Open Frontier Manifesto

Frontier intelligence is the most important technology of our lifetime, and right now a handful of companies own it. They train on all of us: our writing, our code, our questions, our art. Then they lock the result behind an API and call it safety.

We think there's a better way. We think the frontier should belong to the people who built it.

We build in the open, all of it. Not just the weights. We publish the data, the code, the recipes, the training logs and every checkpoint along the way. If you can't reproduce it, it isn't open, and if it isn't open, nobody can check what it does.

We train together. No one of us has a datacenter, but all of us together might. A gaming PC in Osaka, a lab cluster in São Paulo and a spare GPU in a dorm in Lagos all count, and they all add up. Every machine that joins makes the run faster, and every person who joins makes the model theirs too.

We show our work, including the failures. We'll publish every loss spike, every bad run and every dead end, in public, as it happens. We will never fake a number to look bigger than we are. Today we're just getting started: Run 0 is forming, and you can see exactly where it stands.

We share the credit. Every contributor of compute, code, data or evals is named in the ledger. What we build together belongs to everyone, released under licenses that keep it that way.

We do this because we care. Not for an exit or a moat. We do it because we want to understand these systems, because we think humanity's most powerful tools should be inspectable, and because building something enormous with thousands of strangers is one of the best things people can do.

OpenAI was supposed to be this. We'll be it instead.

Frontier intelligence. Built together.

Bring a GPU. Bring an idea. Bring yourself.