Runtimes and hardware

Every function runs on CPU hardware. You choose between CPU Functions, which runs your code from a runtime template, and Custom Container Image, which runs your own Docker image. The runtime you choose decides which languages and frameworks you can use. Both are billed the same way; see Billing.

Hardware choices

• CPU Functions is the general-purpose choice. It suits REST APIs and webhooks, backends for a frontend, microservices, automation and AI agents, and data processing such as CSV and ETL jobs. CPU and container functions can keep dedicated replicas running to avoid cold starts.
• With Custom Container Image, you deploy your own Docker image instead of using a provided runtime. It runs on CPU hardware and is billed as a CPU function. Your image must be built for linux/amd64 and run as a non-root user. By default it must listen on port 8080. To use another port, see Deploy a custom container image.

Language runtimes

Pick the runtime that matches the language and framework you write in. The runtime list in Create Function shows these labels:

Runtime	Code file	Requirements file
Python 3.9 (Http), Python 3.10 (Http), Python 3.11 (Http), Python 3.12 (Http), Python 3.13 (Http), Python 3.14 (Http)	func.py	requirements.txt
Python 3.11 (FastAPI)	func.py	requirements.txt
Python 3.11 (Flask)	func.py	requirements.txt
Node 20.x (Http), Node 22.x (Http), Node 24.x (Http)	index.mjs	package.json
Node 18.x (Express), Node 20.x (Express)	index.js	package.json
Go 1.21 (Http)	handle.go	go.mod
C# .Net 7.0 (Http)	Handler.cs	Function.csproj
PHP 8.2 (Http)	index.php	composer.json
Custom Container Image	Dockerfile	DOCS.md

For Custom Container Image, the second tab is DOCS.md. It sits where the other runtimes show their requirements file and holds the platform’s guidelines for writing the Dockerfile.

Choosing Custom Container Image gives one runtime of the same name. Every other runtime is under CPU Functions.

The Python HTTP runtimes run on Debian Linux, from the official Python slim image. The FastAPI, Flask, and Express runtimes run an ordinary app with its own routes. For how each runtime receives a request, see Requests and responses.

How much CPU each memory size gets

Each memory choice comes with a share of a CPU. The more memory you give a function, the more CPU it gets:

Memory	CPU share
128 MB	0.3 of a CPU
256 MB	0.5 of a CPU
512 MB	0.7 of a CPU
1024 MB	one full CPU

Choosing a runtime

Start with CPU Functions for most functions. Choose Custom Container Image for an application you already run in a container, or for a runtime or library the provided runtimes do not cover.

For Python:

• Python 3.11 (FastAPI) for most new applications: REST APIs, microservices, AI APIs, and apps that need request validation or OpenAPI documentation.
• Python 3.11 (Flask) to move an existing Flask app, for lightweight services, and to use Flask extensions.
• A Python HTTP runtime, such as Python 3.11 (Http), for webhooks and event processing, and to keep cold starts small.

For Node.js, use a Node (Http) runtime, such as Node 24.x (Http), for new functions. Use a Node (Express) runtime, such as Node 20.x (Express), to move an existing Express.js application.

For the runtimes that suit each kind of job, see What you can build.

See Create a function to put a runtime and hardware choice to work, or Deploy a custom container image to use your own image.