For researchers and labs
For researchers and labs.
The compute is rented in Singapore, where the first node, NVIDIA H200 GPUs, is live. It is used by open-weight frontier labs, third-party evaluators, and academic partners. We run evals across the ecosystem, branching into targeted red teaming when failure points are detected.
Launch work
As a partner we help reproduce and diagnose findings, retest, and run safety fine-tuning before launch where a lab wants help fixing what we found. How a campaign runs
The venue
If you run evaluations and keep hitting the compute ceiling, this is the venue you do not have today.
Free · at or below cost
Launch evaluations are free to the lab. Research projects run at or below cost.
Evaluations run against an endpoint the lab hosts. No weights sit on our servers, and the lab sees every query.
The weights are public, so there is no proprietary-weight custody question.
Every job is on a ledger the participants can see: assurance that nobody is using the cluster for capability while you are using it only for safety.
No government money. Private philanthropy is the only funding.
Open weights only
Open weights only, so there is no proprietary-weight custody question.
Gradient updates default-blocked
Gradient updates to the model are blocked by default. Every exception is approved by our technical team, only for a problem we are solving with the lab, and logged on the ledger where the other participants can see it.
Every job on the record
Every job on a public ledger.
No political work
Political work is not permitted. The cluster is not to be used to further the political goals of any party. Any kind of safety testing and evaluation is fine, except work that revolves around politics: anything that enhances censorship, testing around political bias, and the like.
Write to us with the model and what you would run: info@pacificcompute.org, or use the request form.