Runcrate Pricing
Pricing built for the workload.
Not the quota.
Public rate card on every open-source model. Per-second GPU compute, no commit. Dedicated capacity when the rate card starts to hurt. BYOC when compliance demands it.
PLANS
Three plans. No surprises.
Self-Serve
Get an API key. Ship today. Talk to nobody.
$0
per month · pay as you go
What's included
- Pay-per-token inference on every model
- Per-second GPU compute
- OpenAI-compatible endpoint
- Public rate card, no negotiations
- Multi-cloud autoscaling
- Email + community Discord
Deployment
Dedicated
Most popularWhen the rate card starts to hurt.
Custom
volume discounts
What's included
- Everything in Self-Serve
- 40–60% off the public rate card
- Dedicated inference on Arc
- Reserved GPU capacity, no scale-to-zero
- Hands-on engineering setup
- 7-day pilot, parallel to your current provider
- 99.9% uptime SLA
- Slack Connect to our engineers
Deployment
Enterprise
Run it our way. Or run it in your VPC.
Custom
tailored contract
What's included
- Everything in Dedicated
- Custom contract pricing on inference
- BYOC + self-hosted deployments
- Region pinning — US / EU / APAC
- HIPAA-eligible
- SOC 2 Type II datacenter partners
- 99.95% / 99.99% SLA with credits
- Named CSM + on-call engineering
Deployment
INFERENCE · PAY PER TOKEN
The open-source frontier. One API. Public rate card.
No minimum. Stop calling the API, stop paying.
See all 170 modelsINFERENCE · DEDICATED
When the rate card starts to hurt, reserve capacity.
Reserve GPU capacity sized to your traffic, served on Arc — our serving stack that fits 2–3× more tokens per GPU than vanilla vLLM. Commit to a monthly minimum, get a discounted rate on every token. No more 429s on burst, no more TPM ceilings.
Past ~$10k/month, dedicated is cheaper and the bill stops surprising you. We run a 7-day pilot parallel to your current provider — verify cost and latency before flipping a switch.
What you get
- Dedicated capacity sized to your traffic
- 40–60% off the public rate card
- p99 latency floor written into the contract
- Burst-tolerant — no 429s, no TPM ceilings
- 99.9% uptime SLA
- OpenAI-compatible — swap one base URL
COMPUTE · PER SECOND
GPUs by the second.
Stop the instance, stop the meter.
# real-time GPU pricing
GET /api/gpu-pricing
Refreshed every 5 minutes. Hit it from anywhere.
21 GPU SKUs available. Volume + reserved discounts on Dedicated.
TALK TO AN ENGINEER
Tell us what you're shipping.
Self-serve covers most jobs. For dedicated inference, reserved compute, BYOC, or specialty SKUs — tell us your workload. We respond within 24 hours with a rate card and a pilot plan.
Prefer Slack?
Create a Slack Connect channel
FAQ