DOE and its laboratories may need to run a model inside their own infrastructure, preserve a specific version for years, adapt it to a specialized field, and operate without a permanent dependency on an external API.

Open weights support those requirements by allowing an institution to hold and operate the model directly. They do not eliminate the need for permissions, evaluations, sandboxing, audit logs, deployment controls, or disciplined operations. The institution responsible for the work must still design and enforce those controls.

Genesis open models are intended to increase the supply of American open-weight models built for those operating conditions. Its release will include the model weights, a technical report, and public workbench and demonstration artifacts.

“The Genesis Mission exists to turn America’s scientific infrastructure — our laboratories, our data, our computing — into an engine of discovery,” said Darío Gil, DOE Under Secretary for Science and Genesis Mission Director. “Genesis-Science-1 puts that idea to work: an open model trained on the real work of the national laboratories and answerable to the scientists who do it.”

“A country cannot lead in Al if everything it leads in is closed,” said Mark McQuade, co-founder and CEO of Arcee Al. “With Genesis-Science-1, we’re holding American open weights to a demanding standard: useful scientific work under real operating constraints, judged by the people who do it.”

Scientific computing rarely begins with a clean prompt and a single correct answer. A researcher may inherit an aging Fortran codebase, a partially completed simulation campaign, conflicting run logs, and several reasonable options for what to try next. Progress depends on a sequence of technical judgments about which result to trust, which test to run, when to restart, and whether the evidence supports a conclusion.

GS1 will train in scientific workbenches that reproduce these working conditions. Initial areas include high-performance-computing code modernization, experimental analysis, simulation campaigns, materials science, and energy systems.

Each workbench will contain the code, data, tools, documentation, logs, partial results, and failure states needed to reconstruct a research workflow. Training environments may include Python, Fortran, C and C++, MPI and OpenMP, CUDA and HIP, command-line tools, notebooks, simulation packages, and computing schedulers.

The model will operate through a governed execution system. Approved tools will run in sandboxed, staged environments. The system will maintain task state, checkpoint progress, manage retries and recovery, and record the prompts, tool calls, code changes, datasets, intermediate artifacts, and conclusions associated with each run.

Human review will remain part of the process. People will approve decisions involving safety, security, publication, and resource use. GS1 will not receive blanket access to DOE systems.

A successful run must carry a workflow from plan through report, revise the approach when the evidence changes, recover from tool failures, and leave a record that another researcher can inspect. Scientific experts will judge whether the result is sound and whether the supporting evidence is complete enough to reproduce.

Arcee has built and released open-weight models end to end on compressed timelines, with experience spanning data, architecture, pre-training, post-training, evaluation, and deployment. Over six months, the company scaled the Trinity program through increasingly large training runs, culminating in Trinity Large, a 400-billion-parameter sparse mixture-of-experts model. That operating experience allows Arcee to work closely with DOE scientists and contributors throughout development and deliver Genesis-Science-1 on an accelerated schedule.