How payouts work

What you earn

• Serverless: when a job finishes, Ballast adds up the input and output tokens at the model's price. You get 80 percent of that and Ballast keeps 20 percent.
• Dedicated: every healthy rented minute earns you 80 percent of the hourly price divided by 60. A minute only counts if your GPU sent a heartbeat that minute.
• Spot checks are never billed to renters and earn nothing.

The 24 hour hold

New earnings start as Pending. After 24 hours with no failed spot check they become Available. If your machine fails a spot check inside that window, the pending earnings from the last 24 hours are reversed and the ledger shows "Reversed: failed spot check". The hold is how Ballast pays for answers that were real.

Asking for a payout

Open Earnings. You see Available, Pending and Paid out totals and a ledger of every job and rental hour. Once you have $20 of Available earnings, Request test payout turns on. Under $20 the button says how much more you need to reach the $20 minimum. Requesting writes a payout and takes that amount out of Available.

Reading the estimator

On your dashboard the estimator takes your GPU's power draw and your price per kWh (it starts at 450 W and 17 cents) and shows what is left after electricity per busy hour, and per day at 25, 50 and 75 percent busy. It reads from the live price table. It is labelled Estimate, not a promise, because that is exactly what it is.

Here is the honest part. With the starting numbers, a 4090 serving only serverless clears its electricity by about 3 to 5 cents per busy hour. A rented hour in the 24 GB class leaves about 20 cents after power. Those are per busy or rented hour. They say nothing about how many hours renters will ask for, and nobody can promise you that number. Dedicated hours pay much better than serverless, but only when somebody rents them.