Notes ·
The Department of Energy Is Building an American Open Model for Science
The U.S. Department of Energy has launched Genesis Open Models, an effort to develop open-weight foundation models specifically for scientific research.
The first model, Genesis-Science-1, is being built with American open-model company Arcee AI. It will be a trillion-parameter-class model trained with contributions from DOE laboratories and researchers, with its weights and a technical report scheduled for public release later this year.
What makes this notable is the larger open-model landscape.
The United States still has important open models. OpenAI has GPT-OSS under Apache 2.0. Thinking Machines Lab recently released the 975-billion-parameter Inkling with open weights. NVIDIA continues developing Nemotron, Ai2 has OLMo, and Arcee has its Trinity family.
But the center of gravity moved.
Chinese labs including DeepSeek, Alibaba’s Qwen and Moonshot’s Kimi became major sources of capable downloadable models while many leading American labs concentrated on closed systems. Even Arcee explicitly says that this imbalance helped motivate its decision to train American open models from scratch.
For university researchers and national laboratories, provenance matters in ways benchmarks cannot measure.
A research institution may need to keep the same model for years, inspect it, modify it, run it entirely on its own infrastructure and know where its training pipeline came from. Government institutions may also reasonably want an alternative to depending on models developed under another country's laws.
That makes Genesis more interesting than another model release.
It is the U.S. government treating open AI infrastructure as scientific infrastructure.
The goal is not simply to produce an American competitor on a leaderboard. DOE wants researchers contributing scientific data, simulations, software, evaluations and real research environments so the models can become shared tools for fields including energy, materials, climate, biology, fusion and high-energy physics.
That seems like exactly the kind of role national laboratories should play.
Closed frontier models will continue to be useful. But science also needs models researchers can actually possess, reproduce, modify and preserve.