Berkeley Lab supports AI use across research and operations. Start with the Lab’s available AI services and choose the tool that fits your task and the data you will use.
Our core CBorg AI tools are a useful starting point when you need access to one or more supported models for non-sensitive chat, coding assistance, or development. However, a Lab-managed commercial AI subscription may be appropriate when your work requires a specific feature, workflow, team capability, or level of access only available using that product.
ScienceIT can help assess a need for additional capabilities. A paid subscription should have a clear Lab use, an approved funding source, and the required review before purchase.
Current AI Subscriptions
ComputeView is a ScienceIT service that summarizes AI offerings at the Lab and allows staff to check usage, manage budget, and review charges on ongoing projects. These services and agreements will continue to evolve. We will update the available options as they change, with the aim of supporting useful experimentation and being good stewards of the resources that make it possible.
The AI Tools and Access page provides details of what is currently available, how much it costs, how to request access, and which categories of Lab data each service may handle.
ChatGPT Enterprise
For new Lab-funded ChatGPT access, use the centrally managed ChatGPT Enterprise path. We are no longer establishing new individual Pro subscriptions or separate Team/Business workspaces as the standard route for Lab work. IT will work with holders of existing Lab-funded subscriptions on a planned transition. Account holders should wait for guidance before canceling a subscription or merging workspaces so that access, important content, and billing can be reviewed.
Claude Team
Claude Team requests will first be assessed by ScienceIT against the proposed use case and available Lab services. When a paid Claude subscription is recommended, the division and Licensing will coordinate the approved purchase, account ownership, funding, and support arrangements. If Berkeley Lab establishes a Claude Enterprise agreement, Lab-funded individual and team subscriptions will move toward that managed offering through a communicated transition plan.
Comparing LBL-Hosted versus Commercial Models?
LBNL-Hosted Models
Pros:
- Your data is retained within Berkeley Lab systems and networks
- Free to use for building large scale LLM-powered applications
- Supports open research
Cons:
- May be slower to generate responses
- May be occasionally offline during data center maintenance
Commercial Cloud-hosted Models
Pros:
- Highest quality of reasoning, nuanced outputs
- Typically faster to generate
- High-availability and scalability
- Best multi-modal capabilities (speech, vision, document OCR and handwriting recognition)
Cons:
- High cost in large-scale document processing use cases
- Inputs and outputs may be subject to human review for provider policy violations
Contact Us
Unsure which AI route fits your work? Contact the ScienceIT team at scienceIT@lbl.gov. Include what you want to do, what kind of data you will use, whether others need to collaborate, and whether your project has a PID or budget constraint.
For questions about sensitive information, contact the IT Policy team at ITpolicy@lbl.gov.
For purchasing or account changes, follow the current Software and Licensing process. Contact the Licensing team at licenses@lbl.gov for related questions.
