> Markdown copy of https://devshot.co/blog/runpod-pricing (the HTML page is the original; both carry the same facts).
>
> Published: 2026-11-17 · Updated: 2026-11-17
>
> Live software prices and comparisons for AI agents: Toolradar MCP server at https://toolradar.com/api/mcp (setup: https://toolradar.com/for-agents).

# Runpod Pricing 2026: Pods, Serverless and Clusters per GPU Hour

Guide

Runpod pricing in 2026: H100 SXM pods at $3.99 an hour on Secure Cloud or $2.69 on Community, serverless from $0.58 an hour, storage and the fees to watch.

**[Louis Corneloup](https://www.linkedin.com/in/louis-corneloup-0036b5138/)** Founder, Dupple · Updated November 17, 2026 · 11 min read

**Runpod has no plans or seats: you pay per GPU by the second.** An H100 SXM pod costs $3.99 an hour on Secure Cloud or $2.69 on Community Cloud, an A100 SXM $1.79 or $1.39, and an RTX 4090 $0.89 or $0.34. Serverless workers start at $0.58 an hour for 16 GB GPUs. Prices from [Runpod's pricing page](https://www.runpod.io/pricing), updated October 28, 2026 and read November 4, 2026.

The headline rates are honest, but three things move the real bill: which cloud you pick, whether your GPU sits idle, and what happens to your disks when the pod stops. A team that understands those three pays close to the list price. A team that leaves a stopped pod with a big volume disk, or runs an always-on pod for spiky traffic, pays a lot more than it needs to.

**Toolradar data:** Runpod is listed as an alternative to 27 other tools on Toolradar, the software directory we run, which places it #2 of the 44 GPU cloud providers tracked there as of November 3, 2026 ([Runpod on Toolradar](https://toolradar.com/tools/runpod)). Buyers who compare GPU clouds almost always have Runpod on the list. For the product verdict, see our [Runpod review](https://devshot.co/reviews/runpod-review).

## How much does a Runpod GPU cost per hour?

| GPU | VRAM | Secure Cloud, per hour | Community Cloud, per hour |
| --- | --- | --- | --- |
| B300 | 288 GB | $8.99 | $6.94 |
| B200 | 180 GB | $7.99 | $5.98 |
| H200 | 141 GB | $5.29 | $3.59 |
| H100 SXM | 80 GB | $3.99 | $2.69 |
| H100 NVL | 94 GB | $3.19 | $2.59 |
| H100 PCIe | 80 GB | $2.89 | $1.99 |
| RTX Pro 6000 | 96 GB | $2.49 | $1.69 |
| A100 SXM | 80 GB | $1.79 | $1.39 |
| A100 PCIe | 80 GB | $1.79 | $1.19 |
| L40S | 48 GB | $1.09 | $0.79 |
| RTX 6000 Ada | 48 GB | $0.99 | $0.74 |
| A40 | 48 GB | $0.59 | $0.35 |
| RTX 5090 | 32 GB | $1.19 | $0.69 |
| RTX 4090 | 24 GB | $0.89 | $0.34 |
| L4 | 24 GB | $0.59 | $0.44 |
| RTX 3090 | 24 GB | $0.50 | $0.22 |
| RTX A5000 | 24 GB | $0.27 | $0.16 |
| Pro 6000 MIG 48 GB slice | 48 GB | $1.29 | Secure Cloud only |
| Pro 6000 MIG 24 GB slice | 24 GB | $0.69 | Secure Cloud only |

Rates per GPU from Runpod's pricing page (Secure and Community tabs), read November 4, 2026. Runpod's [Pod selection guide](https://docs.runpod.io/pods/choose-a-pod) says the multi-instance GPU slices run on Secure Cloud only.

**Community Cloud is 19% to 62% cheaper than Secure Cloud on the same GPU**, by our arithmetic on the table above, and the gap is widest on consumer cards like the RTX 4090. The guide explains why: Secure Cloud runs in tier 3 and tier 4 data centers with high redundancy, while Community Cloud runs on peer-to-peer hosts with variable reliability. Runpod recommends Secure Cloud for production and sensitive data and Community Cloud for cost-sensitive workloads.

One change to know before you build on Community Cloud: the same guide, last modified October 23, 2026, says **Runpod is no longer accepting new Community Cloud hosts**, while existing Community Cloud resources stay available. Plan for capacity on the cheaper tier to shrink rather than grow, and keep a Secure Cloud fallback in your deploy scripts.

## What does Runpod Serverless cost?

| Serverless GPU class | VRAM | Flex worker, per hour |
| --- | --- | --- |
| B300 | 288 GB | $11.99 |
| B200 | 180 GB | $10.79 |
| H200 | 141 GB | $6.99 |
| H100 (PRO) | 80 GB | $5.39 |
| RTX 6000 Pro (PRO) | 96 GB | $3.49 |
| A100 | 80 GB | $2.72 |
| L40, L40S, 6000 Ada, MIG 48 GB (PRO) | 48 GB | $1.75 |
| 5090 (PRO) | 32 GB | $1.79 |
| 4090 (PRO) | 24 GB | $1.34 |
| A6000, A40 | 48 GB | $1.22 |
| RTX PRO 4500 Blackwell | 32 GB | $1.15 |
| L4, A5000, 3090, MIG 24 GB | 24 GB | $0.75 |
| A4000, A4500, RTX 4000, RTX 2000 | 16 GB | $0.58 |

Flex worker rates from Runpod's pricing page, read November 4, 2026.

**Serverless costs more per hour than a pod but bills only while a worker runs.** Runpod's [Serverless pricing docs](https://docs.runpod.io/serverless/pricing) say flex workers scale to zero when there are no requests and are billed per second, rounded up to the nearest second, from the moment a worker starts until it fully stops. Active workers run around the clock for steady traffic, and their discounts come through a sales inquiry rather than a published rate.

The docs list three billed phases, and two of them surprise people who budget only for execution time. Cold start is billed from worker start until your handler is ready, which mostly means loading the model into GPU memory. After the last job, a worker also waits for new requests for the idle timeout, 5 seconds by default, and that wait is billed too. Pulling the container image happens while the worker is scaled down and is not billed.

## Storage and the fees that are easy to miss

| Storage | While the pod runs | While the pod is stopped |
| --- | --- | --- |
| Container disk | $0.10/GB/month | Not charged, and the data is erased |
| Volume disk | $0.10/GB/month | $0.20/GB/month |
| Network volume, under 1 TB | $0.07/GB/month | $0.07/GB/month |
| Network volume, over 1 TB | $0.05/GB/month | $0.05/GB/month |
| High-performance network storage | $0.14/GB/month | $0.14/GB/month |

Storage rates from Runpod's pricing page and the [Pods pricing docs](https://docs.runpod.io/pods/pricing), read November 4, 2026. Runpod charges no data ingress or egress fees on Pods.

**A stopped pod's volume disk costs twice as much as a running one's.** A 500 GB volume left on a stopped pod costs $100 a month by our arithmetic, for a GPU you are not using. If you stop pods between sessions, move datasets and checkpoints to a network volume, which keeps the same rate whether the pod runs or not.

Four account rules from the Pods pricing docs belong in your runbook:

- **One hour of credit up front.** You need at least one hour's worth of credits for the configuration to deploy an on-demand pod.
- **A $0 balance stops everything.** Pods are stopped automatically when the balance hits $0; pods without a network volume are terminated and their data cannot be recovered. Turn on auto-pay or low-balance alerts.
- **A default spend limit of $80 an hour** across all resources applies until support raises it, which matters before you launch a multi-GPU job.
- **Savings plans are deprecated.** Existing plans run until they expire, but none can be bought, so the published on-demand rates are what a self-serve buyer pays.

## Does Runpod charge per seat?

**No. Neither the pricing page nor the team docs list a per-seat or plan fee.** Runpod's [account docs](https://docs.runpod.io/accounts-billing/manage-accounts) describe converting a personal account into a team account and inviting members with four roles: Basic, Billing, Dev and Admin. A Billing member can see invoices without touching compute, and only Admins can invite people or create clusters, which lets a finance contact watch spend without the power to launch GPUs.

That is different from Modal, a serverless competitor that prices its platform by plan. [Modal's pricing page](https://modal.com/pricing) lists a Starter plan at $0 plus compute with $30 of free credits a month and 3 workspace seats, and a Team plan at $250 a month plus compute with $100 of credits and unlimited seats. For a ten-person team running modest GPU loads, Runpod's lack of a platform fee is a real saving; Modal's credits offset part of its fee.

## What does a fine-tune cost on Runpod?

Take a fine-tune that keeps one H100 SXM busy for 20 hours, with 200 GB of datasets and checkpoints on a network volume for the month. By our arithmetic on the published rates:

| Option | GPU cost, 20 hours | Storage, one month | Total |
| --- | --- | --- | --- |
| Runpod Secure Cloud, H100 SXM | $79.80 | $14 | $93.80 |
| Runpod Community Cloud, H100 SXM | $53.80 | $14 | $67.80 |
| Lambda on-demand, 1x H100 SXM | $85.80 | Instance storage included | $85.80 |
| Modal, H100 SXM5 | $78.98 | See Modal's volume pricing | $78.98 plus any storage |

Lambda's rate is $4.29 per GPU-hour on its single-GPU H100 SXM instance, per [Lambda's pricing page](https://lambda.ai/pricing), plus applicable sales tax. Modal prices an H100 SXM5 at $0.001097 per second; the 20-hour figure is our multiplication. All read November 4, 2026.

**Community Cloud is the cheapest way to run this job, and Secure Cloud costs more than Lambda once storage is added.** The trade-off is reliability: a 20-hour run on a peer-to-peer host should checkpoint often to the network volume, so an interrupted host costs you an hour, not the whole run. If the job needs more than 80 GB of VRAM, the H200 at $5.29 on Secure Cloud is the next step.

## What does a month of inference cost?

Now an inference endpoint for a 7B to 13B model that fits a 24 GB GPU, busy about 4 hours a day in total, or 120 GPU-hours over a 30-day month. We assume 730 hours in an always-on month.

| Option | Monthly cost (our arithmetic) |
| --- | --- |
| Runpod Serverless, 24 GB flex worker, 120 busy hours | $90 |
| Runpod Community Cloud L4 pod, always on | $321.20 |
| Runpod Secure Cloud L4 pod, always on | $430.70 |
| Modal L4, 120 busy hours | $95.90 |

Rates as above; Modal's L4 is $0.000222 per second on its pricing page, read November 4, 2026.

**Serverless wins for spiky traffic by a wide margin.** The always-on Secure Cloud L4 pod only becomes cheaper once the endpoint is busy for about 574 hours a month, or roughly 19 hours a day, by our division of the pod's monthly cost by the serverless hourly rate. Below that, scale-to-zero beats a cheaper hourly rate. Budget a little extra for cold starts and the 5-second idle tail on every burst, which the busy-hours figure does not include.

For larger models the break-even moves earlier. An 80 GB A100 serverless worker costs $2.72 an hour against $1.79 for an A100 SXM pod on Secure Cloud, so a model that needs an A100 and stays busy for more than about 16 hours a day is cheaper on a pod.

## Clusters and enterprise pricing

Runpod's self-serve Instant Clusters list H200 SXM at $6.35 an hour and A100 SXM at $2.15 an hour, with H100 SXM, B200 and L40S by contact with sales, and clusters of up to 64 GPUs with shared storage. Reserved clusters with guaranteed capacity, SLAs and 1 to 12-month terms are priced by sales only. Read on Runpod's pricing page, November 4, 2026.

If you need long reservations, Lambda's 1-Click Clusters publish per-GPU rates for 2-week to 1-year terms, and our [cheapest GPU cloud](https://devshot.co/blog/cheapest-gpu-cloud) comparison lines up on-demand H100 prices across providers.

## Is Runpod the cheapest option for you?

- **Hobby and research runs:** Community Cloud pods are the cheapest per hour on this page. Checkpoint to a network volume and accept the occasional host loss.
- **Production inference with uneven traffic:** Serverless flex workers, with idle timeout and model caching tuned to cut billed cold starts.
- **Steady, high-utilization inference:** a Secure Cloud pod, or active workers if a sales quote beats the pod rate.
- **Teams that want a platform, not a GPU:** compare Modal's plans and our [Modal review](https://devshot.co/reviews/modal-review), or [Replicate](https://devshot.co/reviews/replicate-review) for hosted model APIs.

For the wider field, see our [best serverless platforms](https://devshot.co/reviews/best-serverless-platforms) ranking and the [best AI agent sandboxes](https://devshot.co/reviews/best-ai-agent-sandboxes) guide. If you prefer to run models yourself on a local machine first, the [Ollama review](https://devshot.co/reviews/ollama-review) covers that path, and [Hetzner Cloud pricing](https://devshot.co/blog/hetzner-cloud-pricing) covers cheap CPU servers for everything around the GPU.

Devshot flags GPU price changes like the Community Cloud host freeze as they happen: [subscribe at devshot.co](https://devshot.co/).

## How we priced this

We read Runpod's pricing page (marked updated October 28, 2026) on November 4, 2026, including both cloud tabs for Pods, the Serverless, Clusters and storage tables, plus Runpod's docs on Pods pricing, Serverless pricing, Pod selection and team accounts. Competitor rates come from Lambda's and Modal's pricing pages read the same day. Workload totals, percentages and break-even points are our arithmetic from published rates, before tax; they assume full use of the GPU-hours stated and exclude cold-start time. Devshot does not test products hands-on. Methods are on [how we rate](https://devshot.co/how-we-rate) and corrections go through the [editorial policy](https://devshot.co/editorial-policy).

## FAQ

### How much does Runpod cost in 2026?

Runpod bills per GPU by the second, with no plans or seats. An H100 SXM pod costs $3.99 an hour on Secure Cloud or $2.69 on Community Cloud, an A100 SXM $1.79 or $1.39, and Serverless flex workers run from $0.58 an hour for 16 GB GPUs to $11.99 for a B300, per Runpod's pricing page read November 4, 2026.

### What is the difference between Runpod Secure Cloud and Community Cloud?

Secure Cloud runs in tier 3 and tier 4 data centers with high redundancy, while Community Cloud runs on peer-to-peer hosts with variable reliability and costs 19% to 62% less on the same GPU in the table above. Runpod's docs say it no longer accepts new Community Cloud hosts, though existing capacity remains.

### Does Runpod charge for storage when a pod is stopped?

Yes. A volume disk costs $0.10 per GB a month while the pod runs and $0.20 per GB a month while it is stopped. Container disk is free when stopped but erased, and network volumes keep the same rate either way, from $0.05 to $0.07 per GB a month.

### Is Runpod Serverless cheaper than a pod?

For uneven traffic, yes. A 24 GB serverless worker at $0.75 an hour that is busy 120 hours a month costs $90, against $430.70 for an always-on Secure Cloud L4 pod, by our arithmetic. The pod is cheaper only if the GPU is busy about 19 hours a day or more.

### Does Runpod have a free tier or per-seat pricing?

Runpod's pricing page lists no free tier and no seat fees: you prepay credits and pay for compute and storage. Team accounts with Basic, Billing, Dev and Admin roles are part of the console, and on-demand pods need at least one hour of credit to deploy.

### What happens if my Runpod balance runs out?

Runpod stops pods automatically when the balance reaches $0. Pods with a network volume keep their data; pods without one are terminated and their data cannot be recovered, per Runpod's Pods pricing docs. Auto-pay or low-balance notifications prevent that.

Cite this: Devshot, "Runpod Pricing 2026: Pods, Serverless and Clusters per GPU Hour", November 2026.

Devshot covers the infrastructure behind your stack in one short daily email. [Get it free at devshot.co](https://devshot.co/).

Some offers on this page may be paid placements or contain affiliate links.
