Cloud and platforms
Lambda vs Azure Functions vs Cloud Run Functions: Limits and What Each Bills
In short
The three services differ most in how many requests one instance handles at a time and in what they meter. Lambda event functions run one request per environment and bill requests plus duration. Azure Flex Consumption runs several per instance. Cloud Run functions document up to 1,000 per instance and bill as Cloud Run services.
How many requests does one instance handle at once?
Lambda event functions run one request per execution environment. The Lambda docs say that for each concurrent request Lambda provisions a separate instance of the environment, and a finished environment can be reused for the next request.
Azure Functions on the Flex Consumption plan lets one instance run many executions in parallel. The default HTTP concurrency depends on the instance memory: 4 at 512 MB, 16 at 2,048 MB and 32 at 4,096 MB. Python apps default to 1 on every size. You can set your own value, and it then survives later memory changes.
Cloud Run functions document up to 1,000 concurrent requests per function instance. The 1st generation product handled one.
Azure says that concurrent executions on one instance share memory, CPU and connections. I read the same rule into the Google number. Code that keeps per-request state in a module-level variable is safe in a classic Lambda environment and is a bug on the other two.
Which timeout will you actually hit?
Lambda stops a function at 900 seconds, which is 15 minutes. Lambda Managed Instances functions invoked asynchronously or through most event source mappings can run up to 5,400 seconds.
Azure sets the limit with functionTimeout in host.json. On Flex Consumption the default is 30 minutes and no maximum is enforced. An HTTP-triggered function must still respond within 230 seconds, because of the Azure Load Balancer idle timeout. The legacy Consumption plan defaults to 5 minutes with a 10-minute maximum.
Cloud Run functions allow up to 60 minutes for HTTP-triggered functions. Event-driven functions get 540 seconds, and scheduled or task queue functions get 1,800 seconds.
So the headline numbers hide the shape. Azure looks unbounded and is capped at 230 seconds on the HTTP path. Google gives the longest HTTP window but a shorter one for events. A gateway in front of a function can add another cap; the API Gateway integration timeout page covers one such case.
How large can a request and a response be?
Lambda allows 6 MB each for the request and the response on synchronous calls, and 1 MB for asynchronous calls. A streamed response can reach 200 MB, with bandwidth uncapped for the first 6 MB and 2 MBps after that. If a response is too large, see the Lambda 413 fix page.
Azure lists a maximum request size of 210 MB, the same in every hosting plan.
Cloud Run functions accept 32 MB HTTP requests. Responses can be 32 MB when not streamed and 10 MB when streamed. Eventarc events are limited to 512 KB.
Streaming moves in opposite directions. Lambda raises the response ceiling when you stream, and the Google table lowers it. Check which mode your framework uses before you trust either number.
How much memory and CPU can one instance get?
Lambda memory runs from 128 MB to 10,240 MB in 1 MB steps, and CPU scales with it. At 1,769 MB a function has the equivalent of one vCPU.
Flex Consumption offers three sizes: 512 MB with 0.25 cores, 2,048 MB with 1 core and 4,096 MB with 2 cores. Memory is a menu, not a dial.
Google disagrees with itself here. The comparison page says up to 16 GiB with 4 vCPU for Cloud Run functions. The functions quota page says 32 GiB per instance, and the Cloud Run quota page lists up to 32 GiB and 8 vCPU for services. I would size against the lower figure until Google reconciles them.
Checked against: source 1, source 5, source 7, source 8, source 9
What does each service meter?
All three charge for a call count and for resource time, but the clock starts and stops in different places. The table below lists meters and units only. It carries no prices.
Google says functions operate as Cloud Run services and are billed accordingly. Request-based billing charges CPU and memory only while an instance starts, shuts down or handles at least one request. Instance-based billing charges for the whole life of the instance, with a minimum of 1 minute. Cloud Build and Artifact Registry are priced separately.
Because Cloud Run lets concurrent requests share one instance, I would not compare the three on duration alone. Lambda bills each invocation, so a slow downstream call costs time on every request.
| Meter | Lambda event functions | Azure Flex Consumption | Cloud Run functions |
|---|---|---|---|
| Call counter | Requests | Executions | Requests (request-based billing) |
| Time unit | GB-seconds of duration | GB-seconds of memory while actively executing | vCPU-seconds and GiB-seconds |
| Rounding | Up to the nearest 1 ms | 1,000 ms minimum, then up to the next 100 ms | Up to the nearest 100 ms |
| Warm capacity | Provisioned Concurrency, from enable to disable, rounded up to 5 minutes | Always-ready instances, billed as a baseline | Minimum instances, billed at an idle rate |
| Startup | Init code outside the handler is billed as duration | Not stated on the pages I read | Billed while an instance starts or shuts down |
What should you check before moving a function between them?
Each item below follows from a difference above, so work through them in this order.
- Shared state: does any code assume one request per process? Concurrency of 4 to 32 on Azure and up to 1,000 on Google breaks that.
- HTTP path: add up the gateway, the 230-second Azure response limit and the function timeout.
- Payload: does any response exceed the smallest limit above, once streaming is counted?
- Meters: which of request count, memory time, idle capacity and startup is your largest line item?
- Plan: Microsoft calls the Consumption plan legacy and says Linux Consumption retires on 30 September 2028.
What else do people ask?
Is Cloud Run functions the same product as Cloud Functions?
Google calls Cloud Run functions the latest version of its functions product. Cloud Run functions (1st gen), formerly Cloud Functions (1st gen), has its own limits and its own pricing page.
Does Azure Functions have one limit set?
No. Limits depend on the hosting plan. This page covers Flex Consumption, which Microsoft recommends for new serverless apps, and mentions the legacy Consumption plan where the numbers differ.
Sources
- AWS Lambda quotas, fetched 2026-11-19
- AWS Lambda, Understanding Lambda Event Function scaling, fetched 2026-11-19
- AWS Lambda pricing, fetched 2026-11-19
- Microsoft Learn, Azure Functions scale and hosting, fetched 2026-11-19
- Microsoft Learn, Azure Functions Flex Consumption plan hosting, fetched 2026-11-19
- Microsoft Learn, Concurrency in Azure Functions, fetched 2026-11-19
- Google Cloud, Compare Cloud Run functions, fetched 2026-11-19
- Google Cloud, Cloud Run functions quotas, fetched 2026-11-19
- Google Cloud, Cloud Run quotas and limits, fetched 2026-11-19
- Google Cloud, Cloud Run pricing, fetched 2026-11-19
- Google Cloud, Cloud Run billing settings, fetched 2026-11-19
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. This page compares vendor documents that I fetched on 19 Nov 2026. I did not deploy or run functions on any of the three services for it. It does not cover prices, cold-start timing, regional differences, or Azure Premium and Dedicated plans. Found a mistake? Tell me.