Put a number on it

What does one workflow cost if an agent runs it forever?

Enter your real volume on the left. Everything is an assumption you can change — dial it to your numbers and watch the meter. The point isn’t the default figures; it’s where the two cost curves cross.

Your workflow

$
Build this from tokens
Model (June 2026 rates)
Input tok / call
Output tok / call
Calls / task
Input cached 0%
Cost / request$0.033

A real agent is a loop — set calls/task to 3–6 for a tool-using workflow and watch it climb. Caching cuts input only; it never fixes latency or non-determinism.

$
Back this with a platform

AWS Lambda · $0.20/M requests + ~$1.67/M compute at 512MB / 200ms, x86 (June 2026).

Click through all three. The savings number barely moves — that’s the point. Per-request runtime is rounding error against the agent; the real cost of this path is the build + maintenance below.

$
$
Horizon
Agent tax · 3-yr
$1.10M
paid every request, forever
Compiled · 3-yr
$72,110
build + run + maintain
You keep · over 3 yr
$1.02M
vs. running it live
Live agent meter
Tax accruing at your volume since this page loaded — 0.58 req/sec
$0.00
Cumulative cost
break-even: 46 days
$1.18Mnow18m3yAgentCompiled

Compiling this workflow pays for itself in 46 days, then keeps $1.02M over 3 years — money you're otherwise paying as inference tax on logic that never changes. It also erases about 317 days of cumulative wait time every year.

Assumptions are yours to change. Token and cloud figures are sourced as of June 2026 — verify before relying on them.