RunPod vs Modal: listed GPU prices, September 2026

RunPod Pods cost less per listed H100 GPU-hour than Modal, but Modal’s H100 GPU rate is lower than RunPod Serverless’s.^1,2,1 RunPod lists H100 SXM Pods at $3.49 per hour; Modal lists H100 SXM5 at $0.001097 per second, about $3.95 per hour, on pricing pages checked 28 September 2026.^1,2 That’s a rate comparison: the lower invoice also depends on billed time and additional resources.^2,1

TL;DR

• GPU choice changes the winner.^1,2 Modal’s base GPU charge is lower even against RunPod Pods for B300, B200 and H200 at equal billed duration.^1,2
• RunPod’s displayed Serverless rates are lower than Modal’s GPU rates for L40S and L4; the other shared models below favour Modal on GPU charge alone.^2,1
• An hourly price display doesn’t establish a full-hour minimum.^1,2 RunPod offers hourly and per-second price views; Modal lists GPU-task rates per second.^1,2
• Modal Team costs $250 per month plus compute.² Its Starter plan has no fixed monthly fee, while CPU and memory remain separately priced resources.²

How were the GPU prices compared?

The comparison uses RunPod’s listed Pod and Serverless rates against Modal’s listed GPU-task rates, checked 28 September 2026.^1,2,1 The model treats unlabelled dollar prices as USD, retains original price units and assumes equal billed duration. Currency and tax treatment are not confirmed by those dollar-only displays; obtain both in the quote. The calculation excludes taxes without establishing that every quoted rate is tax-exclusive. Shared models are GPU families listed by both companies.^1,2

RunPod separates dedicated Pods from usage-billed Serverless inference workers.¹ Comparing Modal only with the Pod price would leave out RunPod’s other relevant billing model.^2,1

To compare an hourly rate with a per-second rate, multiply the latter by 3,600; Modal’s listed H100 rate gives roughly $3.95 per GPU-hour.² The rate tables exclude storage, plan fees and Modal’s separate CPU and memory charges.^2,1 Reserved and negotiated prices are outside the comparison.^1,2

For the wider category, see our model infrastructure research. Our research methodology explains the source standard.

Which shared GPU models cost less?

RunPod Pods have the lower listed charge for H100 SXM, RTX PRO 6000, A100 80 GB, L40S and L4; Modal has the lower GPU charge for B300, B200 and H200 at equal billed duration.^1,2 Against RunPod’s displayed Serverless rates, Modal has the lower GPU charge except for L40S and L4.^2,1

GPU family	RunPod Pods, US$/GPU-hour	RunPod Serverless, US$/GPU-hour	Modal GPU tasks, US$/GPU-second
B300	$7.891	$9.981	$0.0019722
B200	$6.791	$8.641	$0.0017362
H200	$4.591	$5.931	$0.0012612
H100	$3.49, SXM1	$4.79, H100 class1	$0.001097, SXM52
RTX PRO 6000	$2.091	$3.491	$0.0008422
A100 80 GB	$1.59, PCIe or SXM1	$2.721	$0.0006942
L40S	$1.091	$1.75, grouped worker class1	$0.0005422
L4	$0.491	$0.69, grouped worker class1	$0.0002222

These are base listed rates before tax, not all-inclusive job prices.² RunPod’s Serverless L40S price applies to a group also containing L40, RTX 6000 Ada and MIG 48GB; its L4 group also contains A5000, 3090 and MIG 24GB.¹ The displayed Serverless quotes don’t identify an active-versus-flex selection, so confirm the worker configuration with the quote.¹

Variant labels also limit the comparison: Modal specifies H200 SXM, while RunPod’s Pod row says H200; its Serverless H100 row leaves the variant unspecified.^1,2,1 RunPod’s price page has Community Cloud and Secure Cloud headings, but the checked text doesn’t assign the displayed Pod prices to either tier.¹

The RunPod SXM Pod quote sits below Modal’s roughly $3.95 hourly equivalent, which sits below RunPod Serverless’s $4.79 hourly quote.^1,2 RunPod also lists H100 PCIe at $2.89 per hour, but that is a different variant from Modal’s SXM5.^1,2 Use the H100 rental-price comparison to examine the wider provider set.

Does per-second billing make Modal cheaper?

Per-second billing alone doesn’t make Modal cheaper than RunPod; the GPU rate and charged duration determine the compute cost together.^1,2 RunPod’s pricing page offers hourly and per-second views, so an hourly display does not establish a full-hour billing minimum.¹ Confirm the selected product’s metering terms in the quote.

RunPod describes Pods as dedicated instances for development and long-running jobs; Serverless bills inference workers based on usage.¹ Modal says it charges actual compute time without charging for idle resources and scales with request volume.²

Confirm which startup, execution and idle periods the selected worker tariff bills. The checked RunPod pricing page does not define those boundaries.¹ Modal’s statement about idle resources also does not establish that the services meter identical portions of a job.²

For a cost test, record charged seconds alongside execution time. Include the worker settings and the GPU variant. The RunPod pricing guide provides the separate product context.

What monthly plan fees and extra charges apply?

Modal Team adds a $250 monthly fee before compute, while RunPod lists no fixed monthly subscription on its checked public pricing page.^2,1 Modal also offers a $0 Starter plan plus compute and an Enterprise plan with custom pricing.² Treat RunPod’s public listing as a price-page comparison, not a guarantee about every enterprise agreement.¹

Plan or public price model	Fixed monthly fee	Listed compute allowance or commercial basis
RunPod public pricing	No fixed subscription listed1	Workload-based charges; reserved capacity and contracts go through sales1
Modal Starter	$0 plus compute2	$30/month free compute2
Modal Team	$250 plus compute2	$100/month free compute2
Modal Enterprise	Custom2	Volume-based discounts2

Starter includes three workspace seats and GPU concurrency of 10; Team lists unlimited seats and GPU concurrency of 50.² The Team fee belongs in a budget when those limits or its additional features are needed. A compute allowance and a fixed plan fee are separate terms; included usage doesn’t turn Team into a free plan.²

Modal separately lists CPU at $0.0000131 per physical core-second and memory at $0.00000222 per GiB-second for GPU tasks.² RunPod shows RAM and vCPU allocations beside its Pod prices, while storage has its own price schedule.¹ Comparing a Pod quote with only Modal’s GPU line leaves out Modal’s other compute charges.^1,2

When should cost favour RunPod or Modal?

RunPod’s H100 Pod rate favours sustained billed GPU time, while Modal’s lower H100 GPU rate against RunPod Serverless favours Modal when comparing equal worker durations before other charges.^1,2,1,2 Changing the GPU can reverse the conclusion: Modal’s B300, B200 and H200 GPU rates undercut even the listed Pod rates.^1,2

Choose the GPU and execution model first. Then compare the same workload using each service’s charged duration, resource charges and applicable monthly fee. The listed price advantage should survive that calculation before it becomes a purchasing decision.^2,1

Record the quoted GPU rate, the time charged and the total cost of completing the job. Test the price advantage against a workload trial and an invoice.

Frequently asked questions

How expensive is Runpod?

RunPod lists H100 PCIe Pods at $2.89 per hour and H100 Serverless at $4.79 per hour on the checked pricing page.¹ Storage can remain a separate cost: its volume disk schedule lists $0.10/GB/month while running and $0.20/GB/month while idle.¹ Include the storage state when budgeting a paused workload.¹

Do I need Modal’s $250 plan to use a GPU?

No: Modal’s Starter plan is $0 plus compute and includes GPU concurrency of 10.² The $250 monthly fee applies to Team, which lists GPU concurrency of 50.² Check whether Starter’s workspace and concurrency limits fit before adding the Team fee to the comparison.

Can region or execution options increase Modal’s rates?

Yes: Modal lists region selection at 1.15-1.75 times base prices and non-preemptible execution at three times base prices.² Apply the relevant multiplier before comparing a configured deployment with a base-price quote.²

Find further comparisons in our published research.