Runpod pricing
Runpod rents graphics cards (GPUs) by the second from a prepaid balance, with no subscription. An RTX 4090 costs $0.34 an hour on Community Cloud and $0.89 on Secure Cloud (Runpod pricing). On our own RTX 4090 Pod, one FLUX ad image used about $0.01 of GPU time, prices checked .
| Plan | Monthly price (USD) | Cost per 1280 x 1280 image (USD) | Paid ads |
|---|---|---|---|
| RTX 4090 Pod | Pay per use | $0.01 | Allowed with conditions |
| RTX 5090 Pod | Pay per use | Not computed | Allowed with conditions |
| L40S Pod | Pay per use | Not computed | Allowed with conditions |
| RTX Pro 6000 Pod | Pay per use | Not computed | Allowed with conditions |
| A100 SXM Pod | Pay per use | Not computed | Allowed with conditions |
| H100 PCIe Pod | Pay per use | Not computed | Allowed with conditions |
| H100 SXM Pod | Pay per use | Not computed | Allowed with conditions |
| Serverless RTX 4090 | Pay per use | Not computed | Allowed with conditions |
| Serverless RTX 5090 | Pay per use | Not computed | Allowed with conditions |
| Serverless H100 | Pay per use | Not computed | Allowed with conditions |
| Storage | Pay per use | Not computed | Allowed with conditions |
| Public Endpoints | Pay per use | Not computed | Allowed with conditions |
Scroll sideways for all 4 columns
What you get on each plan, and notes
- RTX 4090 Pod
- What you get: 24 GB VRAM, 41 GB RAM, 6 vCPUs; Secure Cloud $0.89 an hour, Community Cloud $0.34 an hour, billed by the second
- Notes: Pricing page (updated 8 October 2026) read 24 October 2026; Runpod’s GPU-type API gave the same rates that day.
- RTX 5090 Pod
- What you get: 32 GB VRAM, 35 GB RAM, 9 vCPUs; Secure Cloud $1.19 an hour, Community Cloud $0.69 an hour, billed by the second
- L40S Pod
- What you get: 48 GB VRAM, 94 GB RAM, 16 vCPUs; Secure Cloud $1.09 an hour, Community Cloud $0.79 an hour, billed by the second
- RTX Pro 6000 Pod
- What you get: 96 GB VRAM, 188 GB RAM, 16 vCPUs; Secure Cloud $2.49 an hour, Community Cloud $1.69 an hour, billed by the second
- A100 SXM Pod
- What you get: 80 GB VRAM, 125 GB RAM, 16 vCPUs; Secure Cloud $1.79 an hour, Community Cloud $1.39 an hour, billed by the second
- H100 PCIe Pod
- What you get: 80 GB VRAM, 188 GB RAM, 16 vCPUs; Secure Cloud $2.89 an hour, Community Cloud $1.99 an hour, billed by the second
- H100 SXM Pod
- What you get: 80 GB VRAM, 125 GB RAM, 20 vCPUs; Secure Cloud $3.99 an hour, Community Cloud $2.69 an hour, billed by the second
- Serverless RTX 4090
- What you get: 24 GB 4090 PRO flex worker $1.10 an hour, billed by the second from worker start to full stop; scales to zero between requests
- Notes: Flex rate as shown on the pricing page; active (always-on) worker discounts go through sales.
- Serverless RTX 5090
- What you get: 32 GB 5090 PRO flex worker $1.58 an hour, billed by the second from worker start to full stop; scales to zero between requests
- Notes: Flex rate as shown on the pricing page; active (always-on) worker discounts go through sales.
- Serverless H100
- What you get: 80 GB H100 PRO flex worker $4.79 an hour, billed by the second from worker start to full stop; scales to zero between requests
- Notes: Flex rate as shown on the pricing page; active (always-on) worker discounts go through sales.
- Storage
- What you get: Network volume $0.07 per GB a month under 1 TB and $0.05 over 1 TB, billed hourly; high-performance network storage $0.14 per GB a month; container disk $0.10; volume disk $0.10 while the Pod runs and $0.20 while it is stopped
- Notes: Container disk is erased when a Pod stops; a network volume keeps models and outputs between Pods (Pods pricing docs).
- Public Endpoints
- What you get: Hosted models billed per request, no setup: Wan 2.2 text or image to video at 720p, 5 s, $0.30 a clip; Wan 2.6 5 s $0.50; Seedance 1.0 Pro 5 s at 480p $0.12; FLUX.1 [dev] $0.02 a megapixel; Seedream 4.0 $0.027 an image; Nano Banana Pro Edit $0.14
- Notes: Read on the pricing page’s Public Endpoints table, 24 October 2026. Result URLs expire after 7 days (Public Endpoints docs).
Source: runpod.io/pricing
What one ad costs
Method: $0.01 per image at list price. Measured on our own RTX 4090 Secure Cloud Pod on 17 October 2026: FLUX 3 Klein fast (distilled int8) at 1280 x 1280, 16 steps, made 12 images in 342 s, about 28.5 s each. At $0.89 an hour, 342 s cost $0.085, or $0.007 an image, rounded up to $0.01. GPU time only: Pod start-up, model loading and storage are extra. At the Community Cloud rate of $0.34 an hour the same 342 s cost $0.032.
What you pay per image is the price per hour divided by the images that hour makes. A faster card can cost less per image.
Our own runs on an RTX 4090 Secure Cloud Pod, 16 and 17 October 2026, gave these figures. Each is GPU time only, at the Secure Cloud rate of $0.89 an hour.
| Model and settings | Time per image | GPU cost per image (USD) |
|---|---|---|
| FLUX 3 Klein fast, 1280 x 1280, 16 steps, 12 images in 342 s | 28.5 s | $0.007 |
| FLUX 3 Klein fast, product-in-scene edit, 16 steps | 37 s | $0.009 |
| FLUX 3 Klein full, 1152 x 1440, 50 steps | 4 min | $0.06 |
Scroll sideways for all 3 columns
The fast model is the one to run for ad volume: 12 images for $0.085. At the Community rate, the same 342 seconds would cost $0.032 when a card is free, a figure calculated from the rate. Community Pods cannot attach a network volume, though, so each session downloads its models again on the GPU meter.
The full 50-step model costs $0.06 an image, against $0.007 for the fast one. Save it for the hero image of a campaign.
Runpod GPU prices per hour for images and video
Runpod’s pricing page lists each card at two Pod rates and a Serverless rate. Community Cloud is the cheapest column on every card below. Serverless rates are for the 4090 PRO and 5090 PRO flex workers on those two rows.
| GPU (memory) | Secure Cloud | Community Cloud | Serverless flex worker |
|---|---|---|---|
| RTX 4090 (24 GB) | $0.89 | $0.34 | $1.10 |
| RTX 5090 (32 GB) | $1.19 | $0.69 | $1.58 |
| L40S (48 GB) | $1.09 | $0.79 | $1.75 |
| A100 SXM (80 GB) | $1.79 | $1.39 | $2.72 |
| H100 PCIe (80 GB) | $2.89 | $1.99 | $4.79 |
| H100 SXM (80 GB) | $3.99 | $2.69 | $4.79 |
| RTX PRO 6000 (96 GB) | $2.49 | $1.69 | $3.49 |
Scroll sideways for all 4 columns
Memory decides the card more than speed. Image models of 6 to 17 GB fit the RTX 5090’s 32 GB. A video model of about 40 GB overflows it and spills into slower system memory, which runs slower. The time saved on a 96 GB RTX PRO 6000 can outweigh its higher hourly price.
Community Cloud runs on peer-to-peer hosts whose reliability Runpod’s Choose a Pod page calls “Variable”. Runpod explains the two pools and when each is worth it.
What Runpod storage costs
Storage bills apart from the GPU, and it decides whether your models survive a stop. Runpod has three kinds (Pods pricing):
- Container disk, $0.10 per GB a month, is erased when the Pod stops.
- Volume disk keeps files on that one Pod: $0.10 per GB a month while it runs, $0.20 while it is stopped.
- Network volume, $0.07 per GB a month under 1 TB and $0.05 above it, keeps models and outputs between Pods. It bills hourly whether a Pod runs or not.
For a ComfyUI model library, rent the network volume. It attaches to Secure Cloud Pods only (network volumes). Ours is a 120 GB volume that holds every ComfyUI model we use, for $8.40 a month.
How Runpod billing works
Runpod bills from a balance you load in advance; there is no plan to cancel (Billing docs).
- A Pod bills by the second while it runs. It needs at least one hour’s worth of credits to start.
- A Pod stops when the balance reaches zero. One without a network volume is then deleted, and its files with it (Pods pricing).
- Serverless bills from worker start to full stop, cold starts included, with a 5-second idle timeout by default (Serverless pricing).
- Public Endpoints bill per request. A 5-second 720p Wan 2.2 clip is $0.30.
- Runpod charges nothing for data in or out.
- Credits cannot be refunded or withdrawn, and a default limit of $80 an hour caps spend until support raises it.
Our own account shows how this adds up. October 2026 cost $11.47 to the 24th, $2.21 of it network storage. The Pod share came from 28 short sessions, each stopped when its job ended.
Costs the hourly price leaves out
The per-image figures above count only seconds of generation. Three more things land on the bill.
The first is start-up: a Pod bills while it boots and while ComfyUI loads a model into memory. The second is downloads. A model pulled from Hugging Face on a GPU Pod bills GPU time. A 4-vCPU CPU Pod on the same network volume, on Secure Cloud, does it for $0.12 an hour.
The third is the template’s disk. The official ComfyUI template asks for a 150 GB container disk, billed by the GB and wiped at every stop; 50 GB works.
Is there a Runpod coupon code?
Runpod publishes no coupon code. A new user who signs up with Google through a referral link and loads $10 gets a one-time credit. It runs from $5 to $500 (referral docs, read 24 October 2026).
Users in Europe get a flat $5. Outside Europe, about 96% of new users get $10 or less. Runpod referral code walks through the sign-up.
Where to compare Runpod’s cost
The price tracker puts Runpod beside every app in the database, and cloud GPU prices beside other GPU clouds. The cost per AI ad calculator turns a per-image price into a campaign budget. For setup, how to run ComfyUI on Runpod goes step by step, and the Runpod hub lists every Runpod page.
Questions buyers ask about Runpod pricing
How much does Runpod cost per hour?
For the cards that run image and video models, from $0.34 an hour (RTX 4090, Community Cloud) to $3.99 (H100 SXM, Secure Cloud).
How much is a Runpod H100 per hour?
An H100 PCIe costs $1.99 on Community Cloud and $2.89 on Secure Cloud. A Serverless H100 flex worker costs $4.79 an hour.
Does Runpod charge for a stopped Pod?
Yes, for storage. A stopped Pod’s volume disk costs $0.20 per GB a month, and a network volume bills whether a Pod runs or not.
Is Runpod billed per second?
Yes. Pods bill by the second while they run, and Serverless bills from worker start to full stop, rounded up to the second.
Sources
- Runpod: Runpod pricing, GPU rates per hour for Secure Cloud, Community Cloud and Serverless flex workers, storage rates and the Public Endpoints table; the page says it was updated 8 October 2026; Runpod’s GPU-type API, read with our own account, gave the same rates that day. Checked
- Runpod docs: Pods pricing, billing by the second, the one-hour credit minimum, what happens at a zero balance, disk and network volume rates, no ingress or egress fees. Checked
- Runpod docs: Network volumes, network volumes are only available for Pods in the Secure Cloud. Checked
- Runpod docs: Serverless pricing, flex and active workers, billing from worker start to full stop rounded up to the nearest second, the 5-second idle timeout, billed cold starts. Checked
- Runpod docs: Billing, prepaid credits, non-refundable and not withdrawable; the default spend limit. Checked
- Runpod docs: Referrals, Google sign-up, the minimum first load, the bonus credit by region; the page states no expiry date for the credit. Checked