How Much Does RunPod Cost?
The real price of per-second GPU cloud, card by card, storage and serverless included.
Short answer: RunPod has no free plan. It is pay-as-you-go, with about $10 in credit to start. The cheapest GPU runs at $0.27/hr (RTX A5000) and the RTX 4090 sits at $0.34/hr on Community Cloud. The hourly rate looks tiny, until you run the monthly math: a 4090 left on 24/7 costs around $248, an H100 around $2,110. And storage is billed separately, even while the pod is stopped. We walk through every card, the serverless rates, and what you actually pay for your workload.
Affiliate links: if you sign up through our links, you support us at no extra cost.
RunPod, the key numbers
What each RunPod GPU costs
These are the main on-demand pod cards. The listed price is Community Cloud, the cheaper tier (third-party hosts, more variable availability). Secure Cloud, on RunPod's vetted datacenters, costs more, often close to double: an RTX 4090 goes from $0.34 to $0.69/hr. Everything is billed per second, from pod start to termination.
Prices in USD, Community Cloud tier. Checked June 2026.
RTX A5000 / RTX 4090
The value pick for inference
RTX 4090 24GB at $0.34/hr
- The cheapest entry ticket in the catalog
- Ideal for inference and smaller models
- RTX 4090: $0.34/hr Community, $0.69/hr Secure
- Billed per second, no hourly minimum
RTX A6000 / L40S
48GB for mid-size models
L40S 48GB at $0.86/hr
- 48GB of VRAM for 13B to 34B models
- RTX A6000 at $0.49/hr, A40 at $0.44/hr
- L40 / RTX 5090 around $0.99/hr
- Good VRAM-to-price balance for light fine-tuning
A100 80GB
The training workhorse
A100 SXM 80GB at $1.49/hr
- 80GB of VRAM, the fine-tuning standard
- A100 PCIe at $1.39/hr, SXM at $1.49/hr
- ~$339/month at 8 hrs a day, ~$1,015 at 24/7
- Well under a Lambda A100 ($1.99/hr)
H100 / H200 / B200
Heavy training and LLMs
H200 at $4.39/hr, B200 at $5.89/hr
- H100 PCIe at $2.89/hr, SXM at $3.29/hr
- H200 141GB at $4.39/hr for large LLMs
- B200 180GB at $5.89/hr, the top of the range
- Multi-GPU: the hourly rate scales linearly
Prices checked June 2026 on runpod.io/pricing (Community Cloud tier) and cross-referenced across several sources. Secure Cloud costs more, often close to double (RTX 4090: $0.69/hr confirmed). Secure prices for the other cards should be verified in the deploy console. The RTX A5000 at $0.27/hr was the cheapest GPU in the catalog at the time of checking.
What makes the bill climb
The per-second rate makes RunPod feel almost free. The real story is in the monthly bill. Here are the four line items that actually add up.
The stopped pod that still bills
This is the classic trap. When you stop a pod without terminating it, compute halts, but the storage volume keeps running, and its rate doubles: $0.20/GB/month while stopped versus $0.10/GB/month while running. A 200GB volume left on a stopped pod costs $40/month for nothing.
Storage billed 24/7
Storage is billed independently from compute, around the clock. Container disk and a running volume sit at $0.10/GB/month, network volumes at $0.07/GB/month under 1TB ($0.05 above). It stays modest, but it adds up if you pile up checkpoints and datasets you forget to purge.
Serverless and cold starts
On serverless (flex), you pay per second, from worker start to stop, including the roughly 5-second idle timeout. On bursty traffic, cold starts (container init plus model load) stack up. Active workers avoid cold starts but bill continuously: a tradeoff to weigh against your request volume.
Secure Cloud vs Community
Secure Cloud runs on RunPod's vetted, more reliable datacenters, at close to double the Community rate (RTX 4090: $0.69 versus $0.34/hr). For dev and non-critical inference, Community does the job at half the price. For sensitive production, Secure earns its premium.
- Terminate your pods, don't just stop them: storage doubles to $0.20/GB/month.
- Purge unused volumes and checkpoints: they bill 24/7.
- Bursty traffic? Serverless flex scales to zero, no idle bill.
- Dev or non-critical inference? Community Cloud costs about half of Secure.
- Free egress: pull your models and datasets out with no transfer fees.
How we size the real cost
RunPod's hourly rate does not tell you what you actually pay, because everything depends on uptime. To go from hourly to monthly, we use a 730-hour base (24 hrs x 30.4 days) on Community Cloud, compute only. Here is the breakdown for the most common cards.
- RTX 4090 at 24/7$0.34/hr x 730 hrs, Community Cloud~$248
- A100 80GB at 24/7$1.39/hr x 730 hrs, PCIe~$1,015
- H100 PCIe at 24/7$2.89/hr x 730 hrs, training~$2,110
- Storage 200GB$0.10 running, $0.20 stopped+$20 to $40
Estimates on Community Cloud, compute only unless storage noted. Adjust for your real uptime and the tier you pick.
What you actually pay per month
The price hinges on your card and your uptime. Four typical profiles on Community Cloud, with assumptions stated.
Estimates in USD, Community Cloud. Storage not included unless noted.
Light inference
Serverless A100, ~3 hrs/day
- Serverless flex A100 ($2.736/hr)
- ~91 hrs of compute billed over the month
- Scales to zero between requests, $0 idle
Maker, a 4090 24/7
The most common case
- RTX 4090 Community at $0.34/hr, 24/7
- $248 compute + ~$5 for a 50GB volume
- Roughly half an A100 at the same usage
Serious fine-tuning
A100 80GB, 8 hrs/day
- A100 PCIe at $1.39/hr, ~244 hrs/month
- Great for part-time fine-tuning
- Jumps to ~$1,015 if you leave it on 24/7
Heavy training
H100 PCIe, 24/7
- H100 PCIe at $2.89/hr, full month
- H100 SXM: ~$2,402/month at 24/7
- A 100-hour burst runs ~$289
Estimates on Community Cloud (June 2026), 730-hour base for 24/7, compute only unless storage noted. Adjust for your real uptime. RunPod bills per second: shutting the pod down the moment the job ends is the first cost lever, well before the choice of card.
RunPod's price versus the alternatives
RunPod's entry price next to the other GPU clouds. On consumer cards, RunPod and Vast.ai hold the low end; Lambda and Paperspace target pricier managed infra. Entry prices for comparable cards.
Entry prices checked June 2026. Different billing models.
RunPod
Per second, free egress
- A100 80GB at $1.39/hr, H100 at $2.89/hr
- Billed per second, free data transfer
- Community Cloud about half the Secure rate
Vast.ai
Marketplace, variable prices
- Host marketplace, fluctuating rates
- A100 around $0.50 to $1.00/hr
- Can run cheaper, availability and reliability vary
Lambda
Reliable managed infra
- A100 SXM 80GB at $2.79/hr, H100 from $3.29/hr
- No RTX 4090 in the catalog
- Pricier, but dedicated, stable infra
Entry prices checked June 2026 on the official pages. RunPod and Vast.ai compete on the lowest price, but Vast.ai is a marketplace with variable rates (hence the 'from ~$0.37'). Lambda and Paperspace charge more for managed infra. Modal is serverless per-second like RunPod serverless (A100 ~$2.50/hr). RunPod's real differentiator: free egress, where the hyperscalers charge $0.09 to $0.12/GB on the way out.
So, is RunPod expensive?
Our take after using it: on raw price, RunPod is one of the cheapest around. The real question is usage discipline. Here is when it pays off, and when it stings.
Good value if…
You shut pods down the moment a job ends and stay on Community Cloud for dev. At $0.34/hr for an RTX 4090 and $1.39/hr for an A100, with free egress and per-second billing, RunPod is hard to beat for prototyping, inference and one-off fine-tuning with no commitment.
Less appealing if…
You leave an H100 running 24/7 without watching it. ~$2,110/month arrives fast, and if you forget volumes on stopped pods, storage quietly bills at $0.20/GB/month. For heavy, always-on training, a reserved cluster (contact sales) or a commitment-based rival can come out cheaper.
The verdict
RunPod is an excellent GPU cloud at a fair price, as long as you think per second. Spin up, compute, shut down. Stay on Community for non-critical work, keep Secure for production, and watch your volumes. The card's price matters less than your shutdown discipline.
- Terminate pods the moment a job ends: it is the first cost lever.
- Stay on Community Cloud for dev: about half the Secure rate.
- Watch your volumes: storage bills 24/7, and doubles on stopped pods.
- Irregular traffic? Serverless flex scales to zero between requests.
- Use the free egress to pull models and datasets out at no cost.
Frequently asked questions about RunPod pricing
How much does RunPod cost per hour?
RunPod bills per second, shown as an hourly equivalent. The cheapest GPU is the RTX A5000 at $0.27/hr, and the RTX 4090 runs $0.34/hr on Community Cloud ($0.69/hr on Secure Cloud). 48GB cards like the RTX A6000 sit at $0.49/hr. For training, the A100 80GB is $1.39/hr and the H100 PCIe is $2.89/hr, up to the H200 at $4.39/hr. The listed prices are Community Cloud, the cheaper tier. Secure Cloud, on RunPod's vetted datacenters, often costs close to double.How much does RunPod cost per month at 24/7?
It depends on the card. On a 730-hour monthly base on Community Cloud, an RTX 4090 running 24/7 comes to about $248, an A100 80GB about $1,015 and an H100 PCIe about $2,110 ($2,402 for an H100 SXM). Storage is billed on top, separately. The smart move is not to leave a GPU running around the clock if you do not need it: RunPod bills per second, so shutting the pod down the moment the job ends is the first cost lever.Does RunPod have a free plan?
No, RunPod has no permanent free plan. It is a pay-as-you-go, prepaid-credit model: you add funds (about $10 to start) and every second of GPU time is deducted. Creating an account does not strictly require a card, but you fund before launching a pod. There is a referral program that grants credits once you spend $10, and a startup program that can go up to about $1,000 in credits on application. Be wary of 'free $500' offers on third-party coupon sites, they are not official.What is the price difference between Secure Cloud and Community Cloud?
Community Cloud runs on third-party hosts: it is the cheaper tier, but availability is more variable. Secure Cloud runs on RunPod's vetted, more reliable datacenters, at close to double the rate. Confirmed example: the RTX 4090 goes from $0.34/hr on Community to $0.69/hr on Secure. For dev, prototyping and non-critical inference, Community does the job at half the price. For sensitive production where interruptions are costly, Secure earns its keep. Secure prices for the other cards should be confirmed in the deploy console.How much does storage cost on RunPod?
Storage is billed independently from compute, around the clock. Container disk and a running pod's volume disk are $0.10/GB/month. Watch the trap: on a stopped pod (not terminated), the volume doubles to $0.20/GB/month. Network volumes are cheaper, at $0.07/GB/month under 1TB and $0.05/GB/month above ($0.14 for the high-performance version). So a 200GB volume left on a stopped pod costs $40/month for nothing. The habit: terminate unused pods and purge volumes you no longer need.Does RunPod charge for data transfer (egress)?
No, and it is a real advantage. RunPod charges for neither ingress nor egress: pulling out your models, datasets and results costs nothing in transfer fees. That is a clear difference from hyperscalers like AWS or GCP, which charge egress at $0.09 to $0.12/GB, a line item that climbs fast when you move large volumes. On a workflow where you regularly download checkpoints or trained models, that free transfer is a real saving versus a classic cloud.How does RunPod serverless billing work?
Serverless bills per second, from worker start to full stop, rounded up to the nearest second. Flex workers use the standard rate and scale to zero when there are no requests: you pay nothing while idle. Expect about $2.736/hr for an A100 in flex, or $0.00076/second. Three phases are billed: the cold start (container init and model load), execution, and a roughly 5-second idle timeout. Active workers, pre-warmed, avoid the cold start but bill continuously: a tradeoff to weigh against your traffic volume.Is RunPod cheaper than Vast.ai or Lambda?
It depends on the model. On consumer cards, RunPod ($0.34/hr for the RTX 4090) and Vast.ai (from ~$0.37/hr) play in the same league, right at the bottom of the market. But Vast.ai is a host marketplace with fluctuating rates and variable reliability and availability. Lambda is pricier (A100 from $1.99/hr, H100 from $3.29/hr) but offers more stable managed infra. For inference and one-off fine-tuning at the best price, RunPod is hard to beat, especially with its free egress that few rivals match.How much does it cost to train a model on RunPod?
For training, the cost depends on the card and the duration. On an A100 80GB at $1.39/hr, a 24-hour job comes to about $33. On an H100 PCIe at $2.89/hr, a 100-hour training burst costs about $289. If you chain experiments on an H100 nonstop, expect ~$2,110 for a full month. Since RunPod bills per second, the best saving is to shut the pod down as soon as training finishes and reserve the big cards for runs that truly need them. For inference, an RTX 4090 at $0.34/hr is often enough.Can you cut the RunPod bill with a commitment?
Yes, partly. RunPod offers reserved clusters and long-term commitments, but the rate goes through sales, with no public percentage shown. On serverless, active workers provide the main commitment-style discount (on the order of 20 to 40% per their messaging, not officially itemized). Day to day, the most effective lever is behavioral: shut pods down the moment a job ends, stay on Community Cloud for non-critical work, purge unused volumes and lean on serverless flex scale-to-zero. Shutdown discipline saves more than any discount.
Get our real tool prices in your inbox
Join 2,400+ makers who get our tool tests, pricing teardowns included, every week.
