Last week, we announced a partnership with the U.S. Department of @ENERGY to build Genesis-Science-1 (GS1): an open-weight, trillion-parameter class model built on our Trinity architecture. To make GS1 a reliable tool for discovery, we are accepting data contributions directly from the scientific community.

When we chose to build and train open-weight models in the United States, we were responding to a practical need across government, healthcare, finance, and other critical institutions. These organizations need to own, adapt, and operate the models their systems depend on. Today, Microsoft published a shared set of principles on the role of open-weight models in the broader AI ecosystem. Arcee signed alongside industry peers because those principles reflect a position we have held from the beginning. For sensitive and regulated workloads, institutions need to know what they are deploying and retain control over it. That requires access to the weights, a usable license, a documented training history, and a model supply chain they can account for. A remote API endpoint cannot provide that on its own with. Open weights allow teams to inspect a model, adapt it for specific domains, deploy it inside controlled environments, and keep proprietary data within their own boundaries. The industry is increasingly recognizing a practical reality: durable AI capability depends on systems organizations can audit, reproduce, and operate themselves. We are glad to join Microsoft and our industry peers in supporting open-weight infrastructure. We will continue to back that commitment by building and releasing models that organizations can own and rely on.