AWS Lambda Cost Calculator

AWS Lambda cost is driven by small meters that multiply quickly: requests, billed duration, configured memory, architecture, Provisioned Concurrency, additional ephemeral storage, response streaming, and account-level free-tier allowances. A function can look inexpensive in a single test and still become material when it runs millions of times per month or holds warm capacity for a launch window.

The main compute meter is GB-seconds. Each invocation contributes duration in seconds multiplied by configured memory in gigabytes. Higher memory raises the per-millisecond cost unit, but it also gives the function more CPU allocation and can reduce duration for CPU-bound workloads. That means the cheapest memory setting is not always the smallest memory setting.

Provisioned Concurrency changes the shape of the bill. It can reduce cold-start risk for latency-sensitive traffic, but it adds a capacity charge for configured concurrency during the enabled window. Requests that run during that window also use provisioned-duration rates instead of the ordinary on-demand duration path in this model.

Free-tier treatment needs care because it is an account-level allowance, not a per-function promise. One million monthly requests and 400,000 GB-seconds can be shared across functions and architectures, and other workloads may consume the allowance before this function does. Editable free-tier fields make it possible to model the remaining allowance rather than assuming a fresh account.

A Lambda estimate is strongest when its inputs come from billed duration, memory settings, request counts, and current AWS regional pricing. It is weaker when it ignores nearby services such as API Gateway, event sources, logs, data transfer, storage, retries, and downstream systems.

How to Use This Tool:

Start with a workload shape, then make the rate card match the region and account terms you want to model.

1. Choose a workload preset, architecture, and region profile. Use custom pricing when you need exact regional rates, private terms, or a nonstandard assumption.
2. Enter monthly invocations, average billed duration, and memory allocation. Use billed duration rather than handler-only timing when extensions, initialization, or runtime behavior affect the meter.
3. Choose on-demand, Provisioned Concurrency, or compare mode. When provisioned modeling is enabled, enter configured concurrency, enabled hours, and the share of requests during that window.
4. Set a monthly budget cap if you want the summary and tuning brief to flag the modeled run rate against a planning limit.
5. Open advanced settings for additional ephemeral storage, response streaming bytes, asynchronous event payload request units, free-tier allowances, editable rate fields, and a Compute Savings Plans discount.
6. Review Cost Ledger, Usage Units, Tuning Brief, Lambda Cost Mix, Invocation Cost Curve, and JSON to understand both the total and the drivers.

If a field is outside the accepted range, fix validation before reading the ledger. Memory must stay from 128 MB through 10,240 MB, provisioned hours must stay from 0 through 744, ephemeral storage must stay from 512 MB through 10,240 MB, and percentage shares must remain from 0% through 100%.

Interpreting Results:

Total monthly cost combines request charges, on-demand duration, provisioned duration, provisioned capacity, additional ephemeral storage, modeled response streaming overage, and any entered compute discount. The result excludes taxes, credits not entered in the rate fields, CloudWatch Logs, API Gateway, data transfer, queues, databases, and other surrounding AWS services.

Cost per million invokes is useful for scaling conversations because it normalizes the current workload into a traffic unit. It can still change as free tier is exhausted, provisioned capacity is added, response sizes change, or duration improves with memory tuning.

A low Lambda number can still hide a high application cost if the function causes log ingestion, API requests, database calls, queue messages, NAT gateway traffic, or retries. Use the result as the Lambda slice of a wider architecture estimate.

Technical Details:

Lambda duration billing combines time and memory into GB-seconds. The model splits invocations into on-demand and provisioned-window requests when Provisioned Concurrency is enabled. On-demand GB-seconds can consume the entered free compute allowance. Provisioned-duration GB-seconds and provisioned-capacity GB-seconds are modeled as separate paid meters.

Request charging starts from invocation count, then optionally adjusts asynchronous event sources into request units. In this model, the first 256 KB of an asynchronous event counts as one request unit, and each additional 64 KB chunk up to 1 MB adds another unit. Response streaming is modeled from bytes beyond the first 6 MB per streamed request, after the entered monthly streaming allowance.

Formula Core:

The equations below show the main monthly meters. Rates are editable, so the same usage can produce different totals by architecture, region profile, or custom rate card.

Request cost equals billable request units divided by one million, multiplied by the request rate. On-demand duration cost uses billable on-demand GB-seconds after the entered free compute allowance. Provisioned duration, provisioned capacity, and additional ephemeral storage use their own GB-second rates. The entered compute discount applies to duration and provisioned-capacity charges, not to requests, response streaming, or extra ephemeral storage.

For a default API backend at 25 million monthly requests, 180 ms duration, and 512 MB memory, raw compute is 2.25 million GB-seconds before free-tier compute. Changing memory to 1,024 MB doubles the memory multiplier, but the final cost can still fall if the higher CPU allocation cuts billed duration by more than half.

Limitations and Accuracy Notes:

Public cloud rates change and vary by region, architecture, date, tier, commitment, and account agreement. The built-in rates are planning defaults, and every rate can be edited. Verify the final estimate against the current AWS pricing page, your billing export, Cost Explorer, and any private pricing agreement before using it in a forecast.

The calculation runs from values entered in the browser and does not query your AWS account. It cannot discover real traffic, consumed free tier, regional usage tiers, function settings, or surrounding service costs unless you enter those assumptions yourself.

Worked Examples:

FAQ:

References:

• AWS Lambda pricing
• AWS Lambda memory configuration
• AWS Lambda Provisioned Concurrency