Sign up

Pricing

Pay for runtime only

Billing follows compute time, measured to the second. Capacity follows demand, so the same price model covers a quiet prototype and a busy production service.

Start freeContact us

Compute prices

ComputePrice
  • B200$0.00161 /s
  • H200$0.00112 /s
  • H100$0.000912 /s
  • RTX 6000 Pro$0.000671 /s
  • A100 (80GB)$0.000562 /s
  • A100 (40GB)$0.000531 /s
  • L40S$0.000518 /s
  • A10$0.000294 /s
  • L4$0.000214 /s
  • T4$0.000158 /s
  • CPU only$0.0000062 /vCPU/s
OtherPrice
  • Memory$0.0000021 /GB/s
  • Storage$0.05 /GB/mo

*First 100 GB of storage free

Hobby

For first experiments

Free+ compute / month

  • 3 seats
  • Up to 3 live endpoints
  • 500 containers, 5 GPUs at once
  • Community and chat support
  • 1 day of logs
Start free

Standard

For apps in production

$120+ compute / month

  • Everything in Hobby
  • Unlimited seats
  • Unlimited endpoints
  • 1,000 containers, 30 GPUs at once
  • Custom domains
Start free
Most picked

Enterprise

For teams scaling several models

Custom

  • Everything in Standard
  • Volume pricing
  • No GPU concurrency cap
  • Dedicated Slack channel
  • Guided onboarding
  • ML engineering help
Contact us
Plan comparisonHobbyStandardEnterprise
PriceFree$120Custom
Workspace
ProjectsUnlimitedUnlimitedUnlimited
Live endpoints3UnlimitedUnlimited
Seats310Custom
Custom domains
Data and compliance
Log retention7 days30 daysUnlimited
SOC 2 report
HIPAA, GDPR, ISO 27001
Limits
CPU concurrency5001,000Unlimited
GPU concurrency530Unlimited
Live logs and metrics
Support
Community forum
Shared Slack channel
ML engineering help
Start freeStart freeContact us

Open pricing

Estimate a month from your own workload.

Estimated monthly cost

$0.0000

DetailsEstimated cost
GPU
CPU
Memory
Total
10
seconds
1
1
8 GB

Pricing FAQ

How do per-second prices compare with hourly cloud instances?

Hourly instances bill for boot time, idle time and the headroom you keep for peaks. Our containers start in 1-3 seconds and stop when traffic stops, so the billable time is close to the time spent on requests.

The price also covers orchestration, networking and the routing layer. For uneven traffic the monthly total is usually lower even when the per-second rate is higher.

Show me a billing example

An A10 endpoint with 1 vCPU and 8 GB memory costs $0.000313 per second. At 40,000 requests a month and 1.5 seconds each, that is 60,000 seconds, or $18.78.

Can I use cloud credits?

Yes, on Enterprise plans we can draw on credits you hold with the major clouds. Ask us for the list of supported programs.

Are there discounts for volume or longer terms?

Yes. Committed monthly spend and 12-month terms both lower the per-second rate. Talk to us with your expected usage.

Can I get guaranteed capacity without reserving nodes?

Enterprise plans can include a capacity floor per region. You still pay per second for what runs; the floor only guarantees availability.