Questions & answers
Frequently Asked Questions
Compute, custody, incentives and what we can actually promise.
How long is a launch-evaluation campaign?
We scope campaigns to a three-week target and a four-week maximum from agreed kickoff, including findings delivery and the scoped retest.
Access readiness, staffing and the test matrix are agreed before kickoff. This is the delivery window we plan for, not a claim that every model can receive exhaustive testing in four weeks.
Do you take custody of unreleased weights?
Evaluating models takes mutual trust built over time; we work to build that trust with labs on both sides of the Pacific to launch safer models.
For endpoint-based evaluations, your lab hosts the model and controls access.
Lab-controlled tenancy is designed to keep your engineers in control of the partition, model keys and approved users. We agree configuration, custody and security requirements directly with your lab before work begins.
Could free AI safety compute indirectly fund capability work?
Yes. If we cover work a lab would otherwise pay for, its budget can be spent elsewhere. Safety-only use rules and monitoring do not eliminate that effect, but we believe that researcher incentives will follow compute, and so dedicated AI safety compute means safer models.
What happens when an evaluation finds a serious problem?
We deliver findings privately to the lab’s designated team, help reproduce and diagnose the issue, and offer a scoped retest if the lab makes a change. The lab decides how to respond. Participation carries no obligation to publish findings, and funders or other labs receive no access to results.
Evaluation materials are retained only for the purposes and periods agreed with the lab. Minimal operational records are handled separately.
Is every job public?
No. Access to job metadata and monitoring records is agreed separately from research findings. Findings go privately to the lab and need not be published. We maintain a ledger of cluster activity, telemetry data and just enough information to ensure our cluster is used for safety work.
Why Singapore, and why private philanthropy?
Singapore is a leader in AI safety work, and a practical location for pan-Pacific open model safety collaborations. Our first cluster’s funding is backed by private philanthropic donations to safeguard the open frontier.
Why not give directly to existing evaluators?
Direct grants to good evaluators are valuable. Pacific Compute is complementary: shared compute and practical lab access for work across the open-model ecosystem, with particular attention to open model labs and researchers in Asia.
Is compute available now, and what does it cost?
Our first H200 cluster is rented and live. Launch evaluations are free to the lab; independent safety research runs at or below cost. Allocation depends on the proposed work and available capacity. Current prices are on the Costs page. Send a short request even if you do not yet know your GPU requirements.
Put compute to work.
Bring a model, an evaluation, or the funding to make it happen.