Infrastructure Platform API How it works Research Docs Pricing Try for free

You pay when your agents run

No seats to buy, no monthly minimum, nothing hidden in the invoice. A run is billed for what it actually uses: the tokens it thinks with, the seconds its tools spend working, the models it generates with, and any browser or sandbox time on top.

LLM tokens
$0.20to $50 / 1M

What the agent thinks with. Input and output are priced separately per model, and cached input is a third, roughly 10x cheaper rate - which is what keeps long sessions cheap.

9 agent LLMs →
Tools and connectors
$0.000037per second

Every call to Slack, Notion, GitHub or a CRM, metered only while the call is actually running: $0.133 per hour of tool time.

1500+ actions →
Image and video
$0.02/ image, from $0.20 / clip

One flat catalog price per generation. No per-token math and no resolution surcharge.

500+ models →
Browser and sandbox
$0.0666per sandbox hour

An isolated sandbox for code and media editing, billed per second. A real browser when there is no API, at its session cost plus 15%.

How it is metered →

What a real run costs

four runs off our own account, itemised by the same meter

Search the web and answer

$0.021
2 steps · itemised exactly as the platform billed it
  • GPT-5.6 Luna62k tokens in, 192 out$0.0053
  • Tavily searchone call$0.0160
tokens $0.0053tools and models $0.0160

Draft and send an email

$0.017
2 steps · itemised as billed, tools at today’s rate
  • GPT-5.6 Terra36k tokens in, 194 out$0.0071
  • Mail sendone call$0.0100
tokens $0.0071tools and models $0.0100

Draft and publish a LinkedIn post

$0.019
4 steps · itemised as billed, tools at today’s rate
  • GPT-5.6 Terra104k tokens in, 600 out$0.0191
  • LinkedIn create post2.7 s call$0.0001
tokens $0.0191tools and models $0.0001

Research, then render an image and a voiceover

$0.383
12 steps · itemised as billed, tools and the sandbox at today’s rates
  • GPT-5.6 Terra12 steps of planning$0.1309
  • Image modelone generation$0.1247
  • ElevenLabs voicetext to speech$0.1115
  • Sandbox2.7 s code run$0.00005
  • Tavily searchtwo calls$0.0160
  • Model pick and voicesthree calls, 1.5 s$0.0001
tokens $0.1309tools and models $0.2523

Across our last 938 priced runs the median came out at $0.15: a run that only thinks and answers lands nearer two cents, one that generates media costs what the models cost.

Pay as you go

Metered to the second and the token.

Top up a balance, spend it down. An idle agent costs nothing, and a run can never outspend the cap you set.

Start free →
  • An idle agent costs nothing - there is no per-seat or per-agent fee.
  • Hard caps per agent and per run: a run cannot outspend the ceiling you set.
  • Every run comes back with its own itemised price, tool call by tool call.
Subscriptions

Or prepay a month and get credit on top.

A subscription lands in your wallet as credits every month and is spent down at exactly the rates above. Nothing else changes - same models, same tools, same meter.

+10%at $20
→
+50%at $10,000

Nine tiers, bonus credits growing with the tier. See the whole ladder, or stay on pay as you go - no subscription is required to run anything.

Running at scale? High-volume teams that need custom integrations, data residency, or contract procurement get pricing shaped around how they actually run.
Talk to sales ↗
prepay and save

Prepay a month, take up to +50% in credits.

Credits land in the same wallet and are metered at exactly the rates above. Nine tiers, and the bonus grows with every one of them.

SubscriptionEarly adopter bonusCredits each monthEffective
$20 / mo+$2$22+10%
$50 / mo+$8$58+16%
$100 / mo+$20$120+20%
$200 / mo+$50$250+25%
$500 / mo+$150$650+30%
$1,000 / mo+$350$1,350+35%
$2,000 / mo+$800$2,800+40%
$5,000 / mo+$2,250$7,250+45%
$10,000 / mo+$5,000$15,000+50%

One subscription is active at a time, and you can run without any subscription at all - pay as you go never turns off. Need invoicing, a committed volume or a tier that is not on this list? Talk to us.

the thinking

The models your agents think with.

Priced per 1M tokens and charged on what a run actually consumes. Cached input bills separately at roughly a tenth of the rate, which is why long agent sessions stay cheap. On Claude, GPT-5.6 and GPT-6, input written to the cache bills at 1.25x the input rate, as the providers charge it.

ModelInput / 1MOutput / 1MCached input / 1MContext

Every model above is selectable per agent, and the effort level (low, medium, high) is yours to set. Switching models never rewrites your agent: the instruction is plain text, the model is a setting.

the generation

Every image and video model, one flat price each.

500+ models live on the platform; these are the ones your agents reach for directly. No per-token math, no surprise resolution surcharge.

Prices are the catalog rate per generation unit. Agents pick the model themselves when you do not name one, and recommend_model ranks by price, speed and quality for the job at hand.

the doing

Everything your agent touches, metered by the second.

Connectors, your own MCP servers, a real browser and an isolated sandbox. Charged per second or per call, never per seat.

Built-in tools and connectors

$0.000037 / s

The 1500+ actions in our catalog - Slack, Notion, GitHub, Google, CRMs, payments, scraping. Billed for the seconds the call is actually executing, so a one-second API call costs under a hundredth of a cent. Search, media and voice tools keep their own per-call price. Reading and writing an agent's own files, memory and schedule is free.

and 1500+ more

Your own MCP servers

$0.01 / call

Point the runtime at any MCP server you already run and its tools show up next to ours. Flat base cost per call to your server; whatever that server itself costs stays between you and it.

Browser Use sessions

cost + 15%

A real browser driving a real site when there is no API. Billed as the session's own reported cost plus 15%, charged once when the session finishes - never on the polls in between.

Sandboxes, code blocks and media editing

$0.0666 / h

Agents write and run real code in an isolated sandbox and edit audio and video with ffmpeg, billed per second at $0.0000185. A code run or a media edit is billed for the seconds the call takes; a saved code block only for its execution segment, without start-up, file uploads or failed retries.

the long tail

The wider model platform.

Everything else you can run on our GPUs through the models API, including pipelines we host on request. Same wallet, same metering.

Model name Availability Type Price