Pricing model

Socaity bills on active-only time. You pay for the GPU-seconds your job spends in the PROCESSING state, and nothing for waiting, cold starts, queueing, or failed runs. The mode you pay under depends on the service you call: pay-per-use inference and dedicated hosting.

Mode 1. MaaS pay-per-use

Socaity spins up a GPU on demand, runs your inference, returns the result, and scales the worker back down. The underlying meter is GPU-seconds in the PROCESSING state. Each service surfaces that meter in the unit that makes sense for its workload: per second of inference, per image, or per 1k tokens. A typical job comes in around €0.07 for 3.4 seconds of inference. Cold starts, queue time, and failed runs are not billed. It works like a taxi meter that only runs while you are moving.

Mode 2. Dedicated hosting

You reserve a GPU permanently and pay by the hour for as long as the reservation is live. The GPU stays on whether you send requests or not, so there are no cold starts and no per-call charges. Socaity rents the hardware from an upstream provider (RunPod today; Scaleway and Azure planned) and exposes it through the same SDK and APIPod surface. Closer to renting an apartment: rent is due whether you sleep there or not.

Active-only billing means the meter runs only while the job is in PROCESSING. Everything else is free:

• Cold starts: the seconds spent spinning up a GPU on the first call after idle.
• Queue latency: the time a job spends in QUEUED waiting for a free worker.
• Failed executions: if Socaity infrastructure fails the request, no charge.
• Cancelled jobs: jobs you cancel before they finish processing.
• Idle services: serverless workers scale to zero between calls, so nothing is running and nothing is billed.

A subscription is optional. It layers on top of the two billing modes and adds priority queueing, environment cloning, and enhanced privacy. Subscription tiers do not impose request quotas or rate limits on the SDK or MaaS API; throughput is bounded by GPU worker capacity (cold starts and the number of concurrent warm workers) in the active region. For private serverless, your tier sizes that worker pool.

Your usage breakdown and remaining credit are in the billing dashboard. See Platform → Billing for details.