RunPod pricing per GPU hour: runpod.io, September 2026

RunPod’s cheapest listed GPU Pod is the RTX A5000 at $0.27 an hour. The most expensive is the B300 at $7.89 an hour.¹ These are RunPod’s listed GPU prices from its published pricing page, checked 28 September 2026. The page says it was updated on 27 September 2026.¹ An H100 costs $2.89 to $3.49 an hour depending on the variant, and both A100 variants are listed at $1.59.¹ RunPod bills Pods by the second, so a short job costs a fraction of the hourly rate.²

TL;DR

• RunPod lists 21 GPUs for Pods, from $0.27 an hour for the RTX A5000 to $7.89 an hour for the B300.¹
• The hourly figure is a display rate. Pods are billed by the second, and RunPod says there are no minimums and no ingress or egress fees.²
• Serverless costs more per hour than a Pod on the same GPU: 26% to 71% more for the six GPUs that share a name across both tables.¹ The premium buys scale-to-zero, because RunPod says flex workers scale to zero when idle.³
• Storage is the main extra charge. Volume disk costs $0.10/GB/mo while a Pod runs and $0.20/GB/mo while it sits idle.¹
• Clusters are listed only for the H200 SXM ($4.31/hr) and A100 SXM ($1.79/hr). Every other cluster and every reserved term is “Contact sales”.¹
• The pricing page offers Community Cloud and Secure Cloud views, but the captured table shows one price per GPU.¹ RunPod’s own product page quotes lower “from” prices for four GPUs, so confirm the price in the console before you commit.²

How this ranking was built

The ranking uses one source and one field. The source is RunPod’s GPU Cloud Pricing page at runpod.io/pricing. The page carries the line “Updated September 27, 2026”, and we captured it on 28 September 2026.¹ The ranked field is the hourly price in the Pods table, read in its “Per hour” view. The unit is dollars per GPU per hour, as displayed. The page shows a $ sign but names no currency and gives no tax status. Figures appear here exactly as the page gives them.

Inclusion was every GPU row in the Pods table. The table has 21 rows; all 21 are ranked, with no merges and no exclusions. Ties share a rank and are listed alphabetically. Serverless, Clusters, reserved clusters and Public Endpoints use different units or unpublished prices, so each has its own section below and none is ranked against Pods.

One limit governs the whole table. The Pods table has a Community Cloud / Secure Cloud switch, and the capture records a single price per GPU without showing which view produced it.¹ See our methodology for how Glamdring handles vendor-published figures.

RunPod GPU prices per hour, ranked

The RTX A5000 is the cheapest Pod at $0.27/hr and the B300 is the most expensive at $7.89/hr.¹ VRAM, system RAM and vCPUs are the configuration RunPod lists beside each price.

Rank	GPU	Listed Pod price ($/hr)	VRAM	System RAM	vCPUs
1	RTX A5000	0.27	24 GB	25 GB	9
2	A40	0.49	48 GB	50 GB	9
2	L4	0.49	24 GB	50 GB	12
4	RTX 3090	0.50	24 GB	125 GB	16
5	RTX A6000	0.53	48 GB	50 GB	9
6	Pro 6000 MIG 24GB	0.59	24 GB	31 GB	4
7	RTX 4090	0.74	24 GB	41 GB	6
8	L40	0.82	48 GB	94 GB	8
9	RTX 6000 Ada	0.84	48 GB	167 GB	10
10	RTX 5090	0.99	32 GB	35 GB	9
11	L40S	1.09	48 GB	94 GB	16
11	Pro 6000 MIG 48GB	1.09	48 GB	62 GB	8
13	A100 PCIe	1.59	80 GB	117 GB	8
13	A100 SXM	1.59	80 GB	125 GB	16
15	RTX Pro 6000	2.09	96 GB	188 GB	16
16	H100 PCIe	2.89	80 GB	188 GB	16
17	H100 NVL	3.19	94 GB	94 GB	16
18	H100 SXM	3.49	80 GB	125 GB	20
19	H200	4.59	141 GB	276 GB	24
20	B200	6.79	180 GB	283 GB	28
21	B300	7.89	288 GB HBM3e	251 GB	32

The bundled machine matters as much as the GPU. The RTX 3090 costs a cent an hour more than the L4 but comes with 125 GB of RAM and 16 vCPUs against 50 GB and 12.¹ Both A100 variants cost $1.59, but the SXM row carries twice the vCPUs of the PCIe row.¹

Which RunPod GPU is cheapest per GB of VRAM?

The A40 is cheapest per GB of VRAM, at about $0.0102 per GB per hour. The H100 SXM is the most expensive, at about $0.0436. We divided each listed Pod price by its listed VRAM and rounded to four decimal places.¹

Where a model’s memory footprint decides the card, this view changes the shortlist:

• Lowest: A40 $0.0102, RTX A6000 $0.0110, RTX A5000 $0.0113. Each has 24 GB or 48 GB of VRAM.
• Highest: H100 SXM $0.0436, B200 $0.0377, H100 PCIe $0.0361.
• The B300 costs the most per hour but works out at $0.0274 per GB. That’s less than the H200 ($0.0326), the B200 and every H100 variant.
• The A100s ($0.0199) and the RTX Pro 6000 ($0.0218) are the cheapest routes to 80 GB or more on a single listed GPU.

This is a memory-cost figure, not a throughput figure. A card that is cheaper per GB can still cost more per finished job if it runs the job more slowly. What controls AI inference cost covers the other levers.

How does RunPod bill GPU time?

RunPod bills Pods by the second, from the moment a Pod starts running until you terminate it.² The hourly price is a display unit. RunPod’s Cloud GPUs page says “Billed by the second, never by the hour”, and it says there are no minimums.² The same page’s pricing FAQ says compute is “billed by the millisecond”.² Either way, RunPod’s own example holds: run a job for 3 minutes, pay for 3 minutes.²

Per-second billing helps short, bursty work. It does nothing for a Pod you forget to stop. At the listed rates, a Pod left running for a 730-hour month costs:

GPU	Listed $/hr	730 hours
RTX A5000	0.27	$197.10
RTX 4090	0.74	$540.20
A100 PCIe	1.59	$1,160.70
H100 SXM	3.49	$2,547.70
H200	4.59	$3,350.70
B300	7.89	$5,759.70

Those totals exclude storage. The 730-hour month is our assumption for an always-on Pod, not a RunPod figure.

RunPod also sells spot instances at a discount. They can be evicted when demand spikes, and the page doesn’t publish their prices. RunPod says the console and API show current spot availability and pricing.²

What do Serverless GPUs cost on RunPod?

Serverless runs from $0.58/hr for the 16 GB class to $9.98/hr for the B300.¹ RunPod meters it per second, from worker start to full stop, rounded up to the nearest second.³ Serverless prices a class of GPUs rather than a single card:

Serverless tier	GPUs named	Listed $/hr
280 GB	B300	9.98
180 GB	B200	8.64
141 GB	H200	5.93
80 GB PRO	H100	4.79
96 GB PRO	RTX 6000 Pro	3.49
80 GB	A100	2.72
48 GB PRO	L40, L40S, 6000 Ada, MIG 48GB	1.75
32 GB PRO	5090	1.58
48 GB	A6000, A40	1.22
32 GB	RTX PRO 4500 Blackwell	1.15
24 GB PRO	4090	1.10
24 GB	L4, A5000, 3090, MIG 24GB	0.69
16 GB	A4000, A4500, RTX 4000, RTX 2000	0.58

Six GPUs appear under the same name in both the Pods and Serverless tables. For those six, Serverless costs 26% to 71% more per hour: B300 26%, B200 27%, H200 29%, RTX 4090 49%, RTX 5090 60% and A100 71%.¹ The premium buys autoscaling. RunPod says flex workers scale to zero when idle, so a Serverless endpoint with sparse traffic can cost less over a month than a Pod that runs all the time.³ The decision turns on utilisation. A Pod that is busy most of the hour is cheaper. An endpoint that is idle most of the day favours Serverless.

Serverless also lists GPUs that the Pods table doesn’t: the RTX PRO 4500 Blackwell and the four 16 GB cards.¹ The pricing page has a “Workers” control but doesn’t say whether the rates shown are for flex or active workers.

What do Clusters and reserved capacity cost?

Only two cluster prices are published: H200 SXM at $4.31/hr and A100 SXM at $1.79/hr.¹ The L40S, H100 SXM and B200 clusters are “Contact sales”. RunPod says clusters scale up to 64 GPUs with no commitments.¹

Reserved Clusters list 1, 3, 6, 12 and 12+ month terms for the H200 SXM, A100 SXM, L40S, H100 SXM and B200. Every cell reads “Contact sales”.¹ RunPod describes reserved clusters as offering guaranteed availability, SLA-backed uptime and discounted rates, but it publishes no reserved price.¹ The page doesn’t say whether the cluster figures are per GPU or per cluster, so we have not compared them with Pod rates.

What does RunPod charge for storage and data transfer?

Storage costs $0.05 to $0.20 per GB per month, and RunPod charges no ingress or egress fees.^1,2

Storage type	Listed price
Container disk	$0.10/GB/mo
Volume disk, Pod running	$0.10/GB/mo
Volume disk, Pod idle	$0.20/GB/mo
Network storage (standard), under 1 TB	$0.07/GB/mo
Network storage (standard), over 1 TB	$0.05/GB/mo
Network storage (high-performance)	$0.14/GB/mo

The idle rate is the storage charge to plan for. RunPod says you pay compute only for the time your Pod runs, but an idle Pod’s volume disk doubles in price, from $0.10 to $0.20 per GB per month.^1,2 A 100 GB volume costs $10 a month while the Pod runs and $20 a month while it sits idle. Container disk has no idle price listed.² The page doesn’t say whether the over-1 TB network storage rate applies to the whole volume or only to the portion above 1 TB.

Public Endpoints are a separate product. They are priced per request, per megapixel, per 1,000 characters or per million tokens, depending on the model, not per GPU hour.¹

Why do other RunPod prices disagree with this table?

Other published RunPod figures differ for three reasons: cloud tier, age and third-party copying.

RunPod’s own Cloud GPUs page quotes lower “from” prices for four GPUs than the pricing table shows: A100 PCIe from $1.39/hr, L4 at $0.39, A40 at $0.44 and RTX 3090 at $0.46.² The pricing table lists those GPUs at $1.59, $0.49, $0.49 and $0.50.¹ The H100 PCIe ($2.89) and RTX A5000 ($0.27) match on both pages.^1,2 The most likely explanation is the Community Cloud / Secure Cloud switch, but neither page states that, so treat it as unconfirmed.

Third-party calculators can drift further. One RunPod cost calculator lists the RTX 3090 on demand at $0.19/hr and the A100 SXM at $1.49/hr.⁴ Neither figure appears on RunPod’s pricing page as checked on 28 September 2026.¹

What the ranking cannot tell you

The ranking shows what RunPod lists. It doesn’t show what you will pay. Four gaps remain.

• Tier. The table’s cloud tier is unconfirmed, and RunPod’s product page quotes lower prices for some GPUs.
• Availability. A listed price doesn’t mean capacity is free in your region when you deploy. RunPod says its GPUs run across 30+ regions and doesn’t publish per-region stock on this page.¹
• Spot and reserved rates. Neither is published on the page.
• History. This is our first capture of the page, so we can’t yet say which prices moved since RunPod’s previous update. The next “Updated” date on runpod.io/pricing triggers a refresh.

These are RunPod’s own published figures. They establish what RunPod says it charges, not a market price. For the wider market, see the model infrastructure category and our published research.

Frequently asked questions

Is RunPod free?

No. The pricing page lists no GPU at zero cost, and Pods and Serverless workers bill per second while they run.^1,2 RunPod says you don’t need a credit card to explore the platform, and there are no minimums once you deploy.² Stopped Pods can still incur storage charges.

What is the cheapest GPU to rent on RunPod?

The RTX A5000 is the cheapest listed Pod at $0.27 an hour, with 24 GB of VRAM.¹ For 48 GB, the A40 is the cheapest at $0.49. For 80 GB, both A100 variants are $1.59.¹ On Serverless, the cheapest tier is the 16 GB class at $0.58 an hour.¹

Does RunPod charge for a stopped Pod?

Not for compute, but it does charge for storage. RunPod says you only pay for the exact time your Pod runs.² Volume disk on an idle Pod costs $0.20/GB/mo, double the running rate.¹

Does RunPod publish spot prices?

No. RunPod offers spot instances at a discount, with the risk of eviction when demand spikes, and shows their current availability and price in the console and API rather than on its pricing page.²