Simple, usage-based pricing.

You rent GPU by the second and pay for exactly what your agents use. A budget is a hard ceiling — you can never be surprised by the bill.

Workers per swarm4
GPU-seconds per worker45s

$0.02 / GPU-second, 20% platform fee on metered usage — the real formula behind every run.

Estimated cost

$4.32

Compute · $3.60Platform fee · $0.72

Starter

$0to begin

Everything you need to spawn your first swarm.

Get started
  • Pay only for GPU-seconds used
  • Up to 16 workers per swarm
  • Hard budget ceilings
  • API + MCP access
  • Full audit trail
Most popular

Pay-as-you-go

Meteredper GPU-second

Rent compute by the second. A flat platform fee on top.

Start building
  • Everything in Starter
  • 20% platform fee on metered usage
  • Parallel, sequential & DAG swarms
  • Simulations, schedules & evaluations
  • Signed webhooks & live streaming
  • Scoped API keys with per-key budgets

Enterprise

Customannual

For teams putting production agents behind real budgets.

Contact sales
  • Volume compute discounts
  • SSO / SAML & SCIM
  • Dedicated sandbox capacity
  • Priority support & SLA
  • Custom policies & approvals

The honest comparison

Pay for work, not for waiting.

Every other way to buy agent compute bills you for existence — standby fees, session blocks, idle instances. Swarms bills for seconds worked, under a ceiling you set.

Swarms

metered agent labor

Hosted agent platforms

always-on deployments

Per-task agent products

block-priced sessions

DIY on raw cloud GPUs

build it yourself

Idle cost$0 — agents cost nothing between runs~$155/mo standby per production deployment*Subscription floor even when unused~$3.95/hr per H100 — working or not*
Billing granularityPer GPU-second ($0.02) + 20% feePer node executed + uptime meter~$2.25 per ~15-minute work block*Per instance-hour, always
Overspend protectionHard ceiling — worker stopped at the capSoft alerts; the meter keeps runningBlocks round up; caps are monthly, not per-taskWhatever you build (so: nothing, at first)
Financial audit trailAppend-only ledger, receipt per runTraces, but no money-grade ledgerSession logsWhatever you build
Time to first resultOne API callDeploy pipeline + platform seatsPer-seat product onboardingQueue, sandbox, ledger, budgets — months

*Representative public list prices as of mid-2026 for comparable products and on-demand H100 instances; shapes shown are typical for each category.

$3.61

A six-thread diligence memo

28 minutes, aggregated, with a bear case. An analyst-week costs four figures and arrives Friday.

< $2/mo idle

An overnight ops agent

Wakes on alerts, files the morning report. Pays for seconds awake — sleeping is free.

≈ $9

A 32-persona focus group

Every segment reacts to your pricing page before you ship it. A real panel costs thousands and takes weeks.

GPU-seconds are billed in integer minor units with an append-only ledger and exactly-once charging. Set a budgetUsd on any run and the platform physically cannot exceed it.

Give your agent a workforce.

Start free. Spawn your first swarm in minutes. Pay only for the GPU-seconds you use.