Open weights
We train open-weight models from the ground up and release them under permissive licenses. Run them on your own hardware, fine-tune them, ship them in a product — no usage caps, no gatekeeping, no strings.
Open intelligence
We build, train, and publish open models — weights and all.
Open weights are the floor, not the ceiling. Everything we make is open by default: the models, the recipes and data that trained them, the evals we judge them by, and the tools that serve them. No gated tiers, no waitlists, no "responsible" delay — the moment it's ready, it's yours.
What we open-source
Open science
We publish the work behind every model — data recipes, training code, technical notes, and the evaluations we hold ourselves to — so results are reproducible, not just downloadable. Audit them, or fork them and go further.
Open tooling
The serving, storage, and fine-tuning stack we built for our own models is the same one we hand to you — open source, and live today at pipenetwork.ai.
Every foundational layer of computing — the internet, the browser, Linux — was built in the open. The base layer of intelligence should be no different.
Everything we've published on Hugging Face