Runtime costs 85% less than Lambda MicroVMs for an agent that waits on a model. 100,000 one-minute runs of a 2 vCPU, 4 GiB sandbox, each using 20 CPU-seconds, cost $63.89 a month on Runtime and $420.33 on Lambda MicroVMs. With both CPUs busy the whole time, the same month costs $133.33 on Runtime. Rates checked 25 October 2026.
How Lambda MicroVMs bill
AWS publishes the rates per second: $0.0000276944 per vCPU-second and $0.0000036667 per GB-second. A MicroVM is billed for its baseline, the vCPUs and memory it was created with, whenever it runs. Memory is always 2 GB per vCPU, and the default baseline is 1 vCPU and 2 GB. Usage above the baseline, when a MicroVM scales up for a burst, is billed on top for as long as it lasts. The calculator leaves it out.
| Meter | Lambda MicroVMs, per second | Per hour, for 2 vCPUs and 4 GB | Runtime |
|---|---|---|---|
| CPU | $0.0000276944 per vCPU | $0.1994 | $0.025 per vCPU-hour of CPU used |
| Memory | $0.0000036667 per GB | $0.0528 | $0.0075 per reserved GiB-hour |
On Lambda, the shape costs $0.2522 an hour whether it is busy or waiting. On Runtime, it costs $0.03125 an hour while it waits and $0.08 with both CPUs working.
Example workloads
A month of each, CPU and memory only, from the calculator above. Every shape here is a baseline that Lambda offers.
| Workload | Runtime | AWS Lambda MicroVMs |
|---|---|---|
| Agent runs: 2 vCPU, 4 GiB, 60 s, 20 CPU-seconds, 100,000 a month | $63.89 | $420.33 |
| Tool calls: 1 vCPU, 2 GiB, 2 min, 30 CPU-seconds, 50,000 | $35.42 | $210.17 |
| Long sessions: 4 vCPU, 8 GiB, 30 min, 720 CPU-seconds, 10,000 | $350.00 | $2,522.00 |
| Every CPU busy: 2 vCPU, 4 GiB, 60 s, 120 CPU-seconds, 100,000 | $133.33 | $420.33 |
The saving is 83% to 86% in the first three rows. It is 68% in the last, where the CPUs are never idle.
Where the difference comes from
Two factors add to the gap. Lambda's per-unit rates are higher: about four times Runtime's for CPU and 1.8 times for memory, before counting that a GiB is 7% more than a GB. And Lambda bills the baseline vCPUs whether they work or not, where Runtime bills the CPU time used, with a floor of a twentieth of a vCPU. An agent that spends most of a minute waiting for a model pays Lambda for two whole vCPUs and Runtime for the few seconds it computed.
A MicroVM suspends after an idle time you set and keeps its state for at most eight hours of life. Suspending stops the compute meter, but snapshot storage starts at $0.08 per GB-month, with a one-week minimum. A paused Runtime sandbox costs $0.08 per GB a month too. It keeps its memory while you have credit.
What the calculator leaves out
- Snapshot storage and transfer. Lambda also bills each snapshot read and write, and data transfer at AWS's standard rates.
- Burst above the baseline. Left out on Lambda; on Runtime a sandbox bursts to all its vCPUs and the calculator counts the CPU-seconds you enter.
- Free start. Lambda MicroVMs have no free tier. Runtime gives 100 free hours of a 1 GB machine in the first 30 days, no card.
Price your own runs exactly
After a few runs on Runtime, the CLI prices those same sandboxes at Lambda MicroVMs' published rates:
Terminalruntime compare --from lambda-microvmsIt reads the last 30 days of settled sandboxes (--days up to 90). The full
comparison is Runtime vs AWS Lambda MicroVMs,
AWS Lambda vs an agent sandbox covers
Lambda functions, and the
sandbox cost calculator prices every
provider for your shape.
Sources
- AWS Lambda pricing, checked 25 October 2026
- Lambda MicroVMs
- Runtime pricing
Rival rates on this page were checked on 26 October 2026.