What is Genesis-Science-1?

It’s the first model released under DOE’s Genesis Open Models Initiative — an open-weight AI system built with Arcee AI for scientific computing tasks like materials discovery and fusion research, with a built-in audit trail meant to keep every step reproducible.

Who can apply to contribute to the initiative?

DOE is accepting three types of contributors: groups offering open-weight base models, researchers with scientific datasets, and teams building domain-specific versions of GS1. The pretraining window closes August 14, 2026, and the fine-tuning window closes August 25, 2026, through Argonne National Laboratory’s portal .

Will Genesis-Science-1 be freely available to the public?

Yes — DOE says the initiative is built around open weights and plans to release GS1 with a technical report and demonstration artifacts, though it hasn’t set a public release date beyond the current contributor windows.

US Energy Department Launches Genesis Open Models Initiative

The U.S. Department of Energy (DOE) launched the Genesis Open Models Initiative on August 7, 2026, opening a call for open-weight AI models built specifically for scientific research rather than general-purpose chat or coding. The program sits inside DOE’s broader Genesis Mission, a federal effort to accelerate scientific discovery with AI across the agency’s 17 national laboratories.

Genesis-Science-1

The initiative’s first model, Genesis-Science-1 (GS1), was developed with startup Arcee AI. Unlike a conventional chatbot, GS1 runs inside what Arcee calls a governed execution environment: it carries out approved tools in sandboxed, staged steps, checkpoints its progress, and records every prompt, tool call, code change and dataset it touches. DOE says the goal is reproducibility — a run should leave behind a record “another researcher can inspect” — and the system requires human review before taking safety-, security- or resource-sensitive actions. Target domains include materials discovery, fusion, earth-systems modeling and high-energy physics.

“A country cannot lead in AI if everything it leads in is closed,” Arcee co-founder and CEO Mark McQuade said in announcing the partnership.

Who can apply

DOE is inviting three kinds of contributors: organizations willing to provide open-weight base models, researchers with domain-specific scientific datasets, and teams building domain-adapted versions of GS1. Applications go through Argonne National Laboratory’s contribution portal. The first window, for pretraining contributions, closes August 14, 2026; a second window for fine-tuning efforts closes August 25, 2026, with additional rounds expected roughly every three months.

Part of a wider federal push

The initiative arrives after a year in which US federal agencies have moved unevenly on public AI transparency — the White House recently finalized its own frontier-model review framework but declined to make it public — while private labs have separately courted the research community, as when OpenAI opened a dedicated program for academic researchers. DOE’s approach is explicitly open: both the model weights and the audit trail of how they were produced are meant to be public.

Genesis-Science-1

Who can apply

Part of a wider federal push

Frequently asked questions