How Braise providers earn Compute Credits

Updated September 25, 2026. The public beta is in preparation and can’t be downloaded yet.

Braise connects people who need inference with community machines that run open models. Providers share a supported model on a computer they already have. Users install Braise and connect a supported tool. They don’t need a GPU or a local model runtime.

Share compute, earn compute

In the upcoming free beta, providers will get promotional Compute Credits for each community job their machine completes. They can spend those credits on Braise during the beta. Credits reward finished work, not time spent online, and we don’t guarantee any volume of jobs or reward rate.

You need a platform, runtime and model combination from the provider guide. Sharing only starts when you turn it on. When you turn it off, no new jobs arrive and the ones in progress finish. Electricity, bandwidth and hardware costs are yours. There’s no listing to write and no price to set.

Temporary credits, not money

Beta credits can’t be bought, transferred to someone else, withdrawn or converted into money or cryptocurrency. Every beta balance, granted or earned, goes back to zero before paid V1. None of it becomes a starting balance later on.

If paid V1 clears its own launch conditions, users will be able to buy new prepaid credits. Providers will still earn Compute Credits in V1, and those can’t be cashed out either. A cash reward program may be considered later, as a separate decision. It isn’t a promise tied to the beta, and beta balances won’t be converted into it.

What this beta has to prove

We need to find out whether people finish useful tasks, come back, and find credits useful enough to keep sharing. The starting grant may be enough for early experiments, so a credit balance on its own doesn’t prove that provider rewards matter.

We won’t publish reward rates, performance figures or savings percentages we haven’t measured. Free promotional credits say nothing about the price of a future paid service.

What the provider can read

The machine that runs a request can read the request and its response. Encryption protects data in transit, not on the machine doing the work. Use public or made-up data for your first tests, never confidential information.

A suitable machine has to be online for the model you ask for. If the network can’t serve a request, it returns a capacity error and you’re not charged. It never quietly falls back to a commercial API.

Hear when it opens

The notification form records your interest. It doesn’t reserve a place or create an account. Once the release is approved and published, anyone can sign up with a verified email, with no invitation code and no manual approval.