AWS Lambda Pricing

Lambda charges for two things: each request, and each GB-second the function runs. A GB-second is the memory setting in GB times the billed duration in seconds. In us-east-1 that is $0.20 per million requests and $0.0000166667 per GB-second on x86. A 512 MB function that runs 120 ms, 10 million times a month, costs about $12.

Graviton (Arm) is the easy saving. Its duration rate is about 20% lower and requests cost the same, so most functions get cheaper by roughly a sixth from switching architecture, provided the code and its dependencies build for arm64.

Duration gets cheaper with volume, in tiers. Each tier’s rate covers only the GB-seconds inside it. The first step comes after 6 billion GB-seconds a month on x86, which is about $100,000 of duration in one region. That threshold counts every function in the account.

The free tier is 1 million requests and 400,000 GB-seconds a month, plus 100 GB of response streaming. Every function in the account shares it. The estimator’s default function, 10 million requests at 120 ms and 512 MB, uses all of it on its own. With more than one function in the account, assume it’s spent, which is why the estimator leaves it off.

Provisioned concurrency keeps environments warm. You pay a per-GB-second charge for every second they exist, then a lower duration rate for invocations that run on them. In us-east-1 x86 it beats on-demand only when the environments are busy more than about 60% of the time. Below that, you pay extra for faster start-ups. In AWS GovCloud (US-West) on x86 it never beats on-demand, because the warm charge is larger than the duration saving.

Lambda versus a container comes down to idle time. A 4 GB function that runs a full hour costs more than twice as much as a 2 vCPU, 4 GB Fargate task. The two break even when the function is busy about 41% of the hour. Below that, Lambda is cheaper, because idle time is free.

Managed Instances runs functions on EC2 instances you choose. You pay the EC2 price plus a management fee, which is a median of about 15% of the Linux on-demand price, and requests at the usual rate.

Not every meter is sold everywhere. Lambda@Edge is priced in 22 regions. In us-east-1 its requests cost $0.60 per million, three times the regular rate. The rates table on this page marks what each region lacks.

This page shows on-demand rates only. Compute Savings Plans can discount Lambda duration and provisioned concurrency, and are not in the price list we read.