FOR SCIENTISTS, UNIVERSITIES & OPEN-MODEL DEVELOPERS
A public foundation for American discovery.
Contribute to Liberty AI model development, reproducible research and public-benefit evaluation.
A proposal by AlphaBravo · Authority, funding and deployment remain open.
Discuss research participation
Help develop Liberty AI
American Super Intelligence will give researchers a role in creating, training and evaluating American models. Work will span data preparation, model architecture, pretraining, post-training and adaptation, with reproducible methods and explicit data rights. Liberty Frontier will be the hosted flagship family behind LibertyGPT; Liberty OSS will provide downloadable open-source releases.
Liberty OSS will provide public releases that universities, developers and businesses can examine and build upon. Research partners will contribute methods, evaluations and the knowledge needed to improve those releases.
Research tied to useful capabilities
Super intelligence is the longer-term research horizon. The work begins with questions teams can investigate: how to improve reasoning on unfamiliar problems, support scientific discovery and evaluate increasingly capable systems. Reproducible findings will guide the next step in the program’s long-term vision.
Frontier research will target architectures exceeding two trillion total parameters, with total versus active counts and the architecture established through engineering and evaluation. Parameter count will not substitute for demonstrated reasoning, coding or multimodal performance.
Liberty OSS Code and Liberty Frontier Code will be specialist development tracks, alongside coding in the general models. Future image-generation and voice models will have their own research and evaluation plans and integrate with LibertyGPT and approved applications.
A contribution should identify the task, existing baseline, proposed method and evidence of improvement. Model comparisons will use appropriate U.S.-developed systems. Negative results and limitations belong in the record alongside successful experiments.
Explore the model portfolio and roadmap →
Access for more institutions
the program’s research program will allocate computing capacity for inference, evaluation, model adaptation and selected model-training experiments. These allocations will complement LibertyGPT access and downloadable OSS models; a public chat account will not confer unrestricted research compute.
Compute allocations will have published eligibility and resource limits. Eligibility will include institutions and disciplines without large private computing budgets. Each allocation will require a qualified team and a clear research scope.
Make public releases useful to others
Each Liberty OSS release will describe its license, included development assets, intended uses and evaluation results. Researchers will receive the information needed to understand the release and reproduce documented experiments.
The program will establish the rights required for its open-source commitments. Personal information and material without redistribution rights will still require appropriate handling.
Study the public benefit
Independent teams will examine learning, productivity, entrepreneurship, access and regional economic effects. Studies will include full costs, errors and effects on people whose work changes.
Research methods and findings will be open to scrutiny within applicable data protections. The results will guide how the program develops and expands.