Best Serverless Platform

Google Cloud Functions is Google Cloud's function-as-a-service offering, now presented under the name Cloud Run functions after Google folded its functions product into the Cloud Run platform. Anyone researching the older name will land on the same product; the documentation, quickstarts and pricing all now refer to Cloud Run functions.

The pitch is that you bring the code and Google handles the rest. You write and run small snippets that respond to events, and the operational infrastructure underneath is managed for you. That extends to scale, which Google describes as serving users from zero to planet-scale without thinking about infrastructure, and to orchestration, where functions act as the glue connecting Google Cloud products to one another or to third-party services through events.

Billing is metered to execution time in 100 millisecond increments and you pay nothing while a function is idle, since the service spins instances up and backs them down in response to events. To limit lock-in concerns, functions are built on the open source Functions Framework, which runs the same code across Cloud Run, a local development environment, on-premises and other Knative-based serverless environments.

Published use cases cover integration with third-party APIs and webhooks, such as sending a confirmation email after a Stripe payment or responding to Twilio message events; serverless mobile and IoT backends; real-time file processing, including generating thumbnails immediately after an upload; stream processing driven by Pub/Sub; and video, image and sentiment analysis. Named customers include Commerzbank, Telecom Argentina, Smart Parking and Semios, while HomeAway reported halving development time and lowering cost by more than 66 percent using Cloud Functions with Firestore.

OpenFaaS is a serverless framework for Kubernetes, aimed at teams that want the developer experience of functions without handing their workloads to a public cloud provider. The tagline is serverless on your terms: deploy functions on-premises or in any cloud, as portable OCI images, on any Kubernetes cluster you already run.

The functional model is deliberately unopinionated about language. Official templates cover Node.js, Python, Go, Java, C#, Ruby and PHP, and you can create your own. Just as importantly, existing microservices built with Express.js, Flask, FastAPI, Django or ASP.NET Core can be deployed to OpenFaaS as-is, so adopting it does not require rewriting services into a function shape first. Events can invoke functions from Apache Kafka, AWS SQS and SNS, GCP Pub/Sub, RabbitMQ, PostgreSQL, Cron and MQTT.

Four use cases are called out. Multi-tenant extensibility lets your customers extend your product with custom code, turning snippets into isolated functions built and deployed through a REST API, with each tenant in its own namespace and security boundaries enforced at every layer. Batch and machine learning workloads fan out to thousands of parallel executions with built-in retries. Teams migrating off AWS Lambda to Kubernetes use it to keep the scaling and developer experience they had. And software vendors who must run on their own cloud, a customer's cloud and on-premises get one portable artifact for all three.

The Community Edition is free but limited to personal use, hobbyists and experimentation, explicitly not intended for commercial use. OpenFaaS Pro and Standard add autoscaling, scale-to-zero, event connectors, monitoring dashboards, SSO, RBAC and direct-to-engineering support. Production users named include Waylay, LivePerson, Cognite, RudderStack, Corva and Surge.

Cloudflare Workers is a serverless platform whose defining choice is architectural: rather than running each function in a container, Workers run in isolates. Cloudflare describes isolates as an order of magnitude more lightweight than containers, which is what allows them to scale up and down quickly enough to avoid the cold-start and pre-warming problems that container-based function platforms work around.

That choice produces the platform's two headline claims. The first is placement: a Worker is deployed once and runs in Cloudflare's 335 or more cities by default, with Smart Placement available to run code near your data instead when that minimises end-to-end latency. The second is the billing model. You pay only for CPU time, not for the idle time a function spends waiting on I/O, and there is no pre-provisioned concurrency to buy ahead of a traffic spike.

Workers is intended as a full-stack runtime rather than a place to park isolated functions. React, Vue and Next.js applications ship with their backend logic, databases and storage attached, and bindings connect a Worker to resources such as a database, KV storage, browser rendering, images and Workers AI models with a single declaration. Built-in Queues, Workflows and Cron Triggers cover webhook handling, data processing and scheduled work without a separate scheduler. Common edge patterns including authentication, rate limiting, routing and caching near the user are explicitly supported.

Local development is a first-class concern. workerd, Cloudflare's open source runtime, lets you test changes fully locally before pushing, and language-specific templates give a starting point for a new application. Pricing is tiered from a free plan for personal projects upward.

Azure Functions is Microsoft's event-driven serverless compute service, built so that a team can execute code in response to an event without provisioning or managing the machine that runs it. The product is positioned around writing business logic in the language of your choice and letting triggers and bindings handle the plumbing that connects a function to the rest of Azure.

The service leans heavily on two ideas. The first is triggers and bindings, a declarative way to wire a function to Azure services so that arriving data invokes your code and outbound data is written back without hand-rolled client code. The second is Durable Functions, an extension for stateful orchestration that handles long-running work, multi-step patterns and fault tolerance. Durable Functions performs automatic checkpointing, so an orchestration that fails part way through resumes from its last successful point rather than restarting, which is what makes distributed workflows practical to run on a serverless substrate at all.

Recent positioning is strongly oriented toward AI work. Microsoft describes building agents that integrate with Azure Foundry models and extending agent capability through remote Model Context Protocol servers, alongside more conventional uses such as retrieval augmented generation applications, chatbots, near real-time file and stream processing, and connecting microservices through Azure messaging.

Hosting is plan-based rather than one-size-fits-all. The Flex Consumption plan offers elastic scale with private networking, always-ready instances and concurrency control; an Azure Container Apps plan suits multitype microservices that want consistent networking, observability and billing with the rest of a container estate. Buyers should note a published deprecation: the Linux Consumption hosting plan retires in September 2028, with Flex Consumption named as the migration path. Microsoft was named a Leader in the Forrester Wave for Serverless Development Platforms, Q2 2025.

Netlify Functions is the serverless compute layer of the Netlify platform, designed so that backend logic ships in the same project and the same deploy as the frontend it belongs to. Netlify reports that more than 700 million functions run on the platform every day, covering work such as calling an API, running an AI feature, taking a payment or kicking off a background job.

The product splits into several function types matched to different execution shapes. Standard functions stand up JSON endpoints that a frontend or a third party can call, with the path set in config. Background functions run for up to 15 minutes for work such as AI generation, file processing or data crunching; they respond to the caller immediately and keep working afterwards so the application stays responsive, writing results to Netlify Database, Blobs or a webhook. Scheduled functions take a schedule in config and fire on their own for nightly syncs, digest emails or cleanup, with no scheduler to maintain. Typed event handlers receive webhooks from third-party services or subscribe to platform events such as signups and deploys.

Portability is an explicit design goal. Netlify uses the same module shape as Deno, Bun, Hono and Nitro, so existing code ports with little change. AI work is supported through AI Gateway, which provides access to proprietary and open models from a single function, with results stored in Netlify Database.

Security and rollback are handled at the platform level. Server-side functions can verify identity and gate content before any data leaves, which the browser cannot bypass. Every change gets a preview before it is public and any version can be rolled back in one click. Plans start at a free tier.

Vercel is a deployment and hosting platform for web applications, and its current positioning is unusually specific: the company presents the platform as somewhere to ship apps and agents, and as a target that coding agents themselves can deploy to. Agents work through Vercel's API, CLI, MCP server and Skills, which means an autonomous agent can deploy in what Vercel calls its native language rather than driving a human-facing dashboard.

That agent orientation runs in both directions. Beyond letting agents deploy, Vercel hosts agents as workloads, running them in sandboxed virtual machines with durable backends and access to hundreds of models. The platform also applies agents to its own operations, describing automated agents that investigate errors, plan fixes and open pull requests without a person initiating the work.

Compute runs on Fluid compute, Vercel's execution model for deployed workloads, and the platform includes DDoS mitigation along with traffic and performance insights at every tier. Access control is handled through your existing identity provider, covering internal agents, applications and deployments under one policy rather than a separate set of platform credentials.

Scale references are concrete. Vercel names Notion as powering millions of agent conversations daily on the platform, Zapier as serving over 100 million monthly website visits, and Mintlify as running documentation for more than 20,000 companies. Pricing runs from a free Hobby tier for personal projects through Pro at a fixed monthly rate that includes usage credit, with metered overage on edge requests and data transfer above the included allowances, up to custom Enterprise terms. Included allowances are one million edge requests per month on Hobby and ten million on Pro, after which metered rates apply.