Open AI Research Lab

Building open intelligence the whole field can study.

Quanta is an independent research lab advancing the science of machine intelligence. We publish our models, weights, training recipes, and evaluations in full — because progress that can't be reproduced isn't progress.

47
Papers published
12
Open model releases
1.4M
Weight downloads
100%
Open weights & code
Our mission
We exist to make the foundations of artificial intelligence understandable, reproducible, and shared — so that safety and capability advance together, in the open.
🔓

Open by default

Every release ships with weights, code, data documentation, and evaluation harnesses. No gated artifacts.

🔬

Science, not spectacle

We optimize for reproducible claims and honest baselines over leaderboard theatrics.

🛡️

Safety as a discipline

Interpretability and alignment research run alongside capability work — never as an afterthought.

🌐

For the community

We build infrastructure and tooling that any lab, university, or independent researcher can use freely.

Research directions

Three frontiers we're committed to advancing.

Our agenda is deliberately focused. We go deep on a small number of problems we believe are foundational to the next decade of machine intelligence.

Mechanistic Interpretability

Reverse-engineering the internal circuits of large models — turning opaque weights into legible, testable explanations of behavior.

circuits · features · probing

Efficient Pretraining

Architectures, optimizers, and data curricula that push the compute-quality frontier — strong open models trainable on accessible budgets.

scaling · data · optimization

Alignment & Oversight

Scalable methods for specifying intent, evaluating honesty, and supervising systems that may exceed human ability on narrow tasks.

rlhf · evals · oversight
Careers

Do the most important work of your career, in the open.

We're a small, senior team that values depth over headcount. If you care about rigorous science and shipping work the whole community can build on, we'd love to talk.

See all open roles →