Plans

Start free, grow into Team.

Every plan pays for usage at the rates below, by the second. The free credit renews on the 1st of each month (UTC), is spent before your balance, and does not carry over.

Starter

$0a month

$20 of free credit each month

  • 5 deployed policies
  • 8 GPUs at once
  • 5 benchmarks
  • 600 calls a minute
  • 64 CPU cores · 256 GiB memory · 20 machines
Start free

Team

$200a month

$100 of free credit each month

  • 20 deployed policies
  • 32 GPUs at once
  • 20 benchmarks
  • 2,400 calls a minute
  • 256 CPU cores · 1,024 GiB memory · 80 machines
Upgrade in the console

Custom

Agreedper account

Volume prices and a postpaid credit line

  • Limits set for your fleet
  • Invoiced monthly
Contact us

Pricing

One bill for the whole stack.

Run a policy and benchmark it — all on one key, billed by the second for the GPU, CPU and memory it uses. No GPUs to spin up, nothing to reserve in advance.

Compute
GPU · per card
NVIDIA B200 180 GB$7.50 / hr$0.002083 / sec
NVIDIA H100 80 GB$4.74 / hr$0.001317 / sec
NVIDIA RTX PRO 6000 96 GB Benchmarks$3.64 / hr$0.001011 / sec
NVIDIA A100 80 GB$3.00 / hr$0.000833 / sec
NVIDIA A100 40 GB$2.52 / hr$0.000700 / sec
NVIDIA L4 24 GB Benchmarks$0.96 / hr$0.000267 / sec
CPU and memory · serving a policy
CPU per core$0.0568 / hr$0.00001578 / sec
Memory per GiB$0.0096 / hr$0.00000267 / sec
CPU and memory · checks, evals and benchmarks
CPU per core$0.1703 / hr$0.00004731 / sec
Memory per GiB$0.0289 / hr$0.00000803 / sec

A deployed policy’s machines pay the serving rates. When no serving machine is free, the platform starts one the other way, and it pays the second rates for as long as it runs.

Storage · kept between runs
Weights, compile caches, volumes per GB$0.12 / month

Billed every day it is kept, whether or not anything runs: a deployment’s copy of its weights and its compile cache, and the volumes you upload. Deleting a deployment stops its storage at once.