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.
$0.02 / GPU-second, 20% platform fee on metered usage — the real formula behind every run.
Estimated cost
$4.32
Starter
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
Pay-as-you-go
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
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 granularity | Per GPU-second ($0.02) + 20% fee | Per node executed + uptime meter | ~$2.25 per ~15-minute work block* | Per instance-hour, always |
| Overspend protection | Hard ceiling — worker stopped at the cap | Soft alerts; the meter keeps running | Blocks round up; caps are monthly, not per-task | Whatever you build (so: nothing, at first) |
| Financial audit trail | Append-only ledger, receipt per run | Traces, but no money-grade ledger | Session logs | Whatever you build |
| Time to first result | One API call | Deploy pipeline + platform seats | Per-seat product onboarding | Queue, 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.