DOE launches open-weight models push for science
The US government is betting open-weight AI can become strategic infrastructure — and the first model in its flagship science program is a trillion-parameter bet.
The Department of Energy launched the Genesis Open Models Initiative, a program to build open-weight foundation models for science, with Arcee AI's trillion-parameter Genesis-Science-1 as the first entry. Announced August 7 and hosted at Argonne National Laboratory (genesisopenmodels.anl.gov), the initiative is part of DOE's Genesis Mission — the executive-order-driven effort led by Under Secretary for Science Darío Gil to turbocharge scientific discovery.
DOE is now soliciting contributions from universities, national labs, and companies — data, software, research environments, and evaluations — with the first foundation-stage deadline on August 14 and a post-training window open through August 25. The ask is broad by design: DOE wants the whole pipeline, not just a showcase model.
Arcee, the US open-model lab behind the Trinity family (up to 400B-parameter Trinity Large), says Genesis-Science-1 will be a trillion-parameter-class model released later this year with weights, a technical report, and public demos — and that it has secured the compute to train it. DOE's stated rationale is telling: institutions need to run models on infrastructure they control, keep sensitive work off external APIs, and preserve exact versions for reproducibility. Arcee has pointedly cited DeepSeek, Qwen, and Kimi as proof of how far open weights have advanced in China.
The US government is now explicitly treating open-weight AI as strategic infrastructure, not just a community nicety. The open question is whether a trillion-parameter science model can actually beat the frontier labs at their own game — and whether the national labs can attract the top-tier contributors a program like this needs to avoid becoming a well-funded curiosity. Watch the August deadlines as the first real test of community appetite.
What to watch: whether the August 14 contribution deadline surfaces serious non-Arcee participants — and whether the national labs show up with more than compute.
Would you trust a trillion-parameter government-backed science model more than one from a private lab — or less? Tell us in the comments.
Sources: U.S. Department of Energy · HPCwire · RuntimeWire · Arcee on X
