Render Hosting: Pricing for AI Agents and Backends

Render hosts web services, background workers, cron jobs, Postgres and key-value stores at flat monthly instance prices. For AI products it is the backend: your agent's API, a queue worker for long runs and the database. A 2 GB, 1 CPU web service was $25 a month on 31 October 2026, and billing is prorated to the second. Render does not rent GPUs on the pages we read.

Some links on this page go through /go/ and may earn us a fee; it never changes our order.

Facts at a glance

For other app hosts and GPU options, see the AI hosting overview and the interactive hosting compare.

What Render is for AI work

An agent product needs a place to run code that does not belong in a short function. Render gives you ordinary servers: a web service that listens on a port, a background worker that polls a queue, a cron job, and managed databases. You pay a fixed monthly price per instance, which makes the bill predictable even when a request waits two minutes on a model.

That is the contrast with function platforms. Vercel bills Fluid compute by active CPU and provisioned memory time, and caps function duration. On Render, a request that waits on a model costs nothing extra beyond the instance you already pay for. See Vercel for the other model.

Plans

Render charges for three things, per its pricing page: your workspace plan, a flat subscription; metered features such as bandwidth; and compute.

The pricing page lists the plan names and prices above, with each plan adding to the one before. The FAQ says that with a Pro plan or higher you can enable the Performance build pipeline, which has its own charge for pipeline minutes. A workspace subscription is charged even in a month with no activity, so cancel plans you do not use.

Instance prices

Web service instance types from render.com/pricing on 31 October 2026, per month. The names are CPU and memory.

Render says you can scale up to 12 CPUs and 96 GB of RAM per web service, with SSDs at $0.25 per GB a month. Compute is billed per second and prorated: a web service on a paid plan turned off after 24 hours is billed for 24 hours.

Postgres starts with a free option and paid tiers such as $6 a month for 256 MB, $19 for 1 GB with 100 connections, $40 for 1 CPU and 2 GB, $100 for 2 CPU and 8 GB with 200 connections, and larger up to 128 CPU and 1 TB. Storage is expandable at $0.30 per GB a month, with 1 GB included. Key Value stores start free with 25 MB of RAM, then run $10 a month for 256 MB, $20 for 1 GB and $60 for 3 GB, and scale up to 50 GB of RAM at $750 a month.

What you pay in practice

The configurations are examples we chose, using the table above.

Example one: a small agent API. A 512 MB web service at $7, a 1 GB Postgres at $19, on Hobby at $0 for the workspace. Total $26 a month. Hobby caps bandwidth at 5 GB included, which is fine for a prototype.

Example two: a production agent backend. Pro workspace $25, a 2 GB web service $25, a 2 GB background worker $25 to run long jobs from a queue, a 1 GB Postgres $19. Total $94 a month. Add a 2 CPU, 4 GB web service at $85 instead of the 2 GB one and it is $154.

Example three: a busier service. Pro $25, two 2 CPU, 4 GB web instances at $85 each, one 4 CPU, 16 GB worker at $225, and a 2 CPU, 8 GB Postgres at $100. That is $520 a month. Model tokens are separate: Groq's gpt-oss-120b at $0.15 and $0.60 per million for 40 million input and 8 million output tokens adds about $10.80.

Example four: cost against functions. Vercel's heavy agent example on our Vercel page, 2 million runs with 4 GB held for 60 seconds, came to about $1,480 in usage. On Render the same load, if it fits on a few 4 CPU, 16 GB instances at $225 each, could cost under $1,000, because waiting is free once the instance is paid for. The catch is that you size and scale the instances yourself. These are illustrative figures from list prices, not a tested comparison.

The free tier

Render's free instance docs are blunt: free instances have important limitations and should not be used for production. The ones we read:

• A free web service that goes 15 minutes without inbound traffic spins down, and spins up again on the next HTTP request or new WebSocket connection. That takes about a minute, and browsers see a loading page.
• When a service spins down, changes to its local filesystem are lost.
• Free instance hours reset to 750 at the start of each month and do not roll over. A free web service uses hours while it runs.
• The free Postgres has its own limits. Render's free tier docs say a free database holds 1 GB, only one can be active per workspace, and it expires 30 days after creation.

That makes the free tier a good place for a demo, and a bad place for a customer-facing agent.

Background workers and long jobs

Render's background worker docs describe offloading asynchronous tasks to a separate service listening on a queue. Workers run all the time like web services, but they receive no incoming network traffic. Instead they poll a task queue, such as one backed by a key-value store or a database.

For agents that is the right pattern. The web service accepts a request, writes a job to the queue and returns an id. The worker picks up the job, calls the model, runs tools, saves results to Postgres and notifies the client. The worker has no HTTP request waiting on it, so a long job does not hold a connection open.

Regions

Render says you can deploy services to a list of regions to minimise latency, that it will add regions over time, and that you choose the region when you create a service or datastore. Services that talk to each other should share a region, so that private traffic stays local. Put the backend near the database and near your model provider. The regions docs list Oregon, Ohio, Virginia, Frankfurt and Singapore.

Data and security

The pages we read are about plans and limits, not data processing terms. Render hosts your code and your database, so your data sits on its infrastructure. For regulated data, ask for the security documentation and a data processing agreement. Render's security page lists ISO 27001, SOC 2 Type 2, a GDPR data processing agreement and HIPAA support. As always, keep model API keys in environment variables, never in the repository, and put authentication in front of any endpoint that calls a paid model.

Sizing an agent backend

Agent backends are light on CPU and heavy on waiting, so memory and concurrency decide the instance, not raw compute.

Start small and watch memory. A worker that holds several long conversations in memory grows with concurrency. When it nears its limit, add a second worker before you move to a bigger instance, since two $25 workers give you redundancy as well as capacity, and avoid a single point of failure. Instance prices step up unevenly, as the table above shows, with 2 CPU and 4 GB at $85 costing more than three times the 1 CPU, 2 GB instance at $25. So scaling out is often cheaper than scaling up in the lower tiers.

Render, Vercel and Cloudflare for the same agent app

A reasonable split for a real product: Vercel or Cloudflare for the front end and short routes, Render for the queue worker and database that run long jobs, and a model API or a GPU host for the model. More hosts mean more bills and more places to look when something breaks, so only split when the pieces really differ in shape. For a small product, one Render workspace with a web service, a worker and Postgres is often enough.

Cost habits on Render

1. Delete what you do not use. Instances bill per second while they exist, and a forgotten $225 worker is $225 a month.
2. Use the smallest Postgres that holds your data, and expand storage at $0.30 per GB a month as needed rather than jumping a tier.
3. Watch bandwidth. It is metered beyond the included amount for your workspace.
4. Keep staging on Hobby or small instances, and turn it off when no one is testing.
5. Review instance size quarterly. Many services sit at 20 percent memory use on an instance that was sized for a launch.
6. Pair it with a cheap model route for early tiers. gpt-oss-20b at $0.20 and $0.30 per million tokens on Cloudflare costs $10.40 for our example 40 million in and 8 million out.

Leaving Render

Render runs standard processes, so leaving is not hard. Your web service is a container or a build from a repository, and the Postgres is standard Postgres, which you can dump and restore elsewhere. Keep configuration in files in the repository and secrets in a manager you control. The harder part is the surrounding setup: private networking between services, cron schedules and environment groups, which you will recreate on the new host. Note that Render does not offer full migration professional services for moving into it, per its FAQ, though it lists partners and migration credits.

Who should pick Render

Alternatives

• Vercel: functions and previews, Pro at $20 a month with $20 of credit.
• Cloudflare Workers: Free plan with 100,000 requests a day, Paid from $5 a month; models alongside on Cloudflare Workers AI.
• Modal: serverless Python with GPUs, for the model side.
• Runpod and Lambda Cloud: rented GPUs.
• AWS Bedrock: models inside AWS.

Getting started

1. Connect a repository and create a web service. Start on Hobby with a small instance.
2. Add Postgres in the same region.
3. Add a background worker for long jobs, and a queue.
4. Move to a Pro workspace before real customers.
5. Set environment variables for keys, and put authentication on the API.
6. Watch usage for bandwidth, which is metered, and for instance sizes you can drop.

See how to build an AI agent for the application side, and vLLM if you plan to pair it with your own model server.

Our take

Render is the plain choice for an agent backend: flat prices, normal servers, workers for long jobs and managed Postgres. Its weakness is that it is not an AI platform, so the model sits elsewhere, and scaling is your job. For short, spiky agent calls, a function host may cost less. For long runs and steady traffic, flat instances usually win. The free tier is for demos only. Our conclusions come from Render's pricing page and docs, and we deployed nothing for this page.

Related pages

• Hosting hub and AI hosting overview
• Private AI hub
• How to self-host an LLM