Summary: Under its Genesis Mission initiative, the US Department of Energy is creating an open-weight foundation model family, with Genesis-Science-1 as the first model. Arcee AI leads development while DOE and participating national laboratories supply reviewed scientific materials, tasks, and evaluations through a contribution portal. First-round applications close August 14, 2026.
Under its Genesis Mission initiative, the U.S. Department of Energy is creating an open-weight foundation model family whose purpose is to accelerate scientific discovery. Genesis-Science-1 (GS1) is the first member, made together with Arcee AI. A contribution portal hosted by DOE and operated by Argonne National Laboratory went live at the same time, inviting the research community to supply data, tasks, and evaluations; first-round applications close August 14, 2026.
It is unusual for a national government to fund open-weight foundation models itself. The signal carries more weight than any single model.
Genesis-Science-1: Arcee Leads on the First Model
Responsibilities are split cleanly. The main development line belongs to Arcee AI: obtaining compute, curating training data, carrying out pretraining and post-training, standing up the governed execution environment, and taking evaluation and release through to the end. DOE and the participating national laboratories, meanwhile, anchor the system in genuine scientific practice — providing reviewed scientific materials, specifying representative research tasks, designing evaluations, and checking what the system produces.
Arcee calls GS1 an open-weight model paired with a governed research system: the point is not to answer questions but to carry scientific computing workflows through from start to finish, leaving behind a reproducible record of the work. Arcee CEO Mark McQuade was blunt about the stakes: “A country cannot lead in AI if everything it leads in is closed.” GS1, he added, is held to a demanding standard — useful scientific work under real operating constraints, judged by the people who actually do the science.
A National Contribution Portal: Data, Tasks, Evaluations
Universities, national laboratories, companies, scientific nonprofits, and research organizations may all take part, with two contribution tracks planned for 2026. Foundation-stage contributions are looking for material suitable for pretraining, midtraining, or context extension: scientific text, code, documentation, and structured technical collections. The post-training track is far more specific — expert demonstrations, annotated task examples, research software, datasets, complete workflow environments, reinforcement-learning tasks, held-out evaluations, scoring rubrics, tests, and verifiers. Scientists and engineers are also being recruited to review tasks, model behavior, and completed artifacts.
Applying is deliberately lightweight. The public form asks only for descriptions and metadata; no scientific material changes hands at that stage. Five gates decide selection — scientific fit, rights and handling, expert and evaluation readiness, technical integration, and final program selection. Those who pass gain early evaluation access while the system is still under development, along with credit in the technical report and release materials.
What stands out is how concrete the scope is for a government program. Rather than asking for generic “AI data,” the portal asks scientists for the artifacts of real work: experimental and observational data from user facilities, simulation outputs and run logs from supercomputing campaigns, materials and chemistry collections, and the tools researchers use to conduct and review their work. Materials enter only after the Department’s own release-review process is finished, so provenance is settled up front rather than litigated afterward. The August 14 deadline covers the first contribution window; infrastructure providers can submit rolling inquiries about training and evaluation capacity.
From Executive Order to National Open Models
The Genesis Mission was set in motion by executive order in November 2025 as a Department of Energy-led initiative, with a stated goal of doubling the productivity and impact of American science and engineering within a decade. It links DOE’s 17 national laboratories, leadership-class computing facilities, and partners from industry and academia, and Under Secretary for Science Darío Gil directs it. The target domains for GS1 reach across materials discovery, energy systems, earth systems modeling, fusion, biology, and high-energy physics. Three principles — open science, reproducibility, and responsible AI — form the official framing, with the models positioned as a shared public asset.
What It Means for Open Source and Science
Two things follow. On the open-source side, this amounts to a major government treating open weights as an instrument of national science and technology strategy — a marker worth noting as the policy debate over American open models continues, and one that Arcee’s quote captures in a sentence. On the science side, the opening is tangible: experimental data, simulation outputs, and research software that seldom find their way into training material now have an official route into a foundation model, and the governance terms are declared before contribution rather than negotiated after release. DOE’s strict release-review process, meanwhile, rules out a “dump the weights online” style of open source. How many institutions actually file before the August 14 deadline is the first thing to watch.