Review

Runpod Review 2026: Pricing, Pros, Cons and Alternatives

What Runpod's Community and Secure pods, flex and active serverless workers and storage cost, where each one is the cheaper choice, and how Lambda, Modal, Replicate and Vast.ai compare.

Product links may be affiliate links. How we rate 4 alternatives covered
TL;DR

Runpod is worth it in 2026 for developers who want an H100 or A100 by the hour without a contract. An H100 PCIe pod costs $1.99 an hour on Community Cloud or $2.89 on Secure Cloud, and an A100 PCIe $1.19 or $1.79, per Runpod's pricing page (updated October 28, 2026, read November 4).

The real cost depends on the product. Pods bill while they run, serverless flex workers bill only while they work but at a higher hourly rate, and a stopped pod still pays for its disk.

Key factsBest for: Developers who want a specific GPU by the second, from…
  • Updated: November 15, 2026
  • Best for: Developers who want a specific GPU by the second, from consumer cards to H200 and B300
  • Price as of November 15, 2026: H100 PCIe Pod $1.99/hr Community or $2.89/hr Secure; serverless H100 flex $5.39/hr.
  • A per-second GPU cloud with pods, serverless workers and clusters, cheapest on its Community column.
  • Cheapest pod: RTX A5000 at $0.16 an hour on Community Cloud
  • A100 80GB pod: From $1.19 an hour (PCIe, Community Cloud)
  • Billing: Per second for pods and serverless, no data ingress or egress fees on pods
  • Default spend limit: $80 an hour across all resources, raised by support

Pros

  • The cheapest fixed-price H100 in this comparison, on Community Cloud, with no contract
  • No data ingress or egress fees on Pods
  • Flex serverless workers scale to zero between requests

Cons

  • Community Cloud runs on peer-to-peer hosts with variable reliability, and Runpod no longer adds new hosts
  • Pods without a network volume are terminated, with data lost, when the balance hits $0
  • Active serverless worker discounts are not published, only offered through sales
Cheapest podRTX A5000 at $0.16 an hour on Community Cloud
A100 80GB podFrom $1.19 an hour (PCIe, Community Cloud)
BillingPer second for pods and serverless, no data ingress or egress fees on pods
Default spend limit$80 an hour across all resources, raised by support

Runpod sells GPU time three ways: Pods (a dedicated GPU you start and stop), Serverless (inference endpoints that scale workers up and down), and Clusters for multi-node jobs. All three bill by the second, and the pricing page lists more than 20 GPU types, from an RTX A5000 at $0.16 an hour to a B300 at $8.99. For a developer, the question is less "is Runpod cheap" than which of its products fits the job.

Community or Secure Cloud is the first choice. Runpod's docs describe Community Cloud as peer-to-peer providers with variable reliability, and Secure Cloud as T3 and T4 data centers with high redundancy, recommended for production and sensitive data. The docs also say Runpod no longer accepts new Community Cloud hosts, so the cheaper column may not grow.

Toolradar data: Toolradar tracks 44 GPU cloud tools as of November 3, 2026, and Runpod is listed as an alternative to 27 of them, second of 44 in the category. Replicate is the most cross-shopped, listed as an alternative to 40. For hourly prices across eight providers, see our cheapest GPU cloud comparison.

What is Runpod?

Runpod is a GPU cloud for AI work: fine-tuning, batch jobs, notebooks and model serving. A Pod is a container on a dedicated GPU with a container disk, an optional volume disk and an optional network volume that can move between Pods. Serverless turns a container into an endpoint whose workers start on demand; Public Endpoints sell ready-made models such as FLUX.1 and Whisper per request.

It is not a managed model API. If you only need text completions, a hosted LLM API is simpler; our best LLM APIs ranking and the free LLM APIs list cover that route. Runpod makes sense when you need your own weights, your own container or a specific GPU.

How Runpod works

Pods bill per second for compute and storage while running, with no data ingress or egress fees. You need at least one hour of credit for the configuration to deploy an on-demand Pod. When a Pod stops, the GPU charge ends but the volume disk keeps billing, at $0.20 per GB a month instead of $0.10 while running. A 200 GB volume parked for a month costs $40, by our arithmetic.

The balance rule can cost you data. Pods stop when the account balance hits $0. Runpod's Pod pricing docs say Pods with a network volume keep their data, while Pods without one are terminated and their data cannot be recovered. Turn on auto-pay or low-balance alerts before a long training run.

Serverless bills each worker from start until it fully stops, rounded up to the second, including cold start and the idle timeout (5 seconds by default) after its last job. Flex workers scale to zero and pay the listed rate; active workers run 24/7, with discounts available only through sales, per the Serverless pricing docs.

Runpod key features

Two price columns for Pods
Every Pod GPU has a Community Cloud and a Secure Cloud rate. The gap is widest on consumer cards: an RTX 4090 is $0.34 an hour on Community against $0.89 on Secure, while A100 SXM is $1.39 against $1.79.
Flex serverless workers
Flex workers scale to zero, so an idle endpoint costs nothing beyond storage. The trade is the rate: a flex worker costs more per hour than the same GPU as a Pod, nearly twice as much for an H100.
Clusters without a contract
Instant Clusters scale to 64 GPUs with shared storage and per-second billing: H200 SXM at $6.35 and A100 SXM at $2.15 per GPU-hour. H100 SXM, B200 and L40S clusters, and all reserved clusters, are priced by sales.
Storage priced per GB
Container disk is $0.10 per GB a month, network storage $0.07 under 1 TB and $0.05 above, and high-performance network storage $0.14. Network volumes are the safe place for weights and checkpoints.

Runpod pricing

Prices were read on November 4, 2026 on Runpod's pricing page (marked updated October 28, 2026), Lambda pricing, Modal pricing, Replicate pricing and Vast.ai's live price list. All are on-demand list rates in US dollars.

One H100 for 100 hours a month (a fine-tuning month), by our arithmetic: Runpod Community $199, Vast.ai about $285 at the H100 PCIe median listing of $2.85, Runpod Secure $289, Lambda $329 for a single H100 PCIe, Modal about $395 for GPU time alone at $0.001097 a second, and Replicate $549. Vast.ai is a marketplace, so its figure moves daily.

Pod or serverless? A flex H100 worker costs $5.39 an hour against $2.89 for a Secure H100 Pod, so serverless is cheaper while the endpoint is busy less than about 54% of the day, roughly 13 hours, by our arithmetic. For A100s the line sits near 66%, about 16 hours. An endpoint serving traffic two hours a day costs about $323 a month on H100 flex workers, against $2,110 for a Secure H100 Pod left on for 730 hours.

Savings plans are deprecated and can no longer be bought; reserved capacity and serverless reservations go through sales. The default spend limit is $80 an hour across all resources until support raises it.

Methodology: list prices from each vendor's own pricing page and docs, read on November 4, 2026. Monthly totals and break-even points are our arithmetic on those prices; they exclude storage and, for Modal, CPU and memory, which it bills separately.

PlanPriceBest for
Runpod H100 PCIe Pod$1.99/hr Community, $2.89/hr Secure80 GB VRAM, 188 GB RAM, 16 vCPUs, billed per second
Runpod H100 SXM Pod$2.69/hr Community, $3.99/hr Secure80 GB VRAM, 125 GB RAM, 20 vCPUs
Runpod A100 PCIe Pod$1.19/hr Community, $1.79/hr Secure80 GB VRAM, 117 GB RAM, 8 vCPUs
Runpod RTX 4090 Pod$0.34/hr Community, $0.89/hr Secure24 GB VRAM, 41 GB RAM, 6 vCPUs
Runpod Serverless H100 (flex)$5.39/hrScales to zero; active workers run 24/7 with discounts through sales
Runpod Serverless A100 (flex)$2.72/hr80 GB, billed per second including cold start and idle timeout
Runpod storage$0.10/GB/mo disk; $0.07/GB/mo network (<1 TB)Volume disk on a stopped Pod rises to $0.20 per GB a month
Lambda 1x H100 PCIe$3.29/GPU/hrSelf-serve instance, plus applicable sales tax
Modal H100$0.001097/secServerless, scales to zero; CPU and memory billed separately; $30 monthly credit on Starter
Replicate H100$0.001525/sec ($5.49/hr)Per-second hardware billing for models you run on Replicate
Vast.ai H100 PCIeFrom $2.53/hr, median $2.85/hrLive marketplace listings; prices change daily

Runpod pros and cons

What we like

  • The cheapest fixed-price H100 in this comparison, on Community Cloud, with no contract
  • No data ingress or egress fees on Pods
  • Flex serverless workers scale to zero between requests
  • Instant Clusters scale to 64 GPUs without a commitment

What could be better

  • Community Cloud runs on peer-to-peer hosts with variable reliability, and Runpod no longer adds new hosts
  • Pods without a network volume are terminated, with data lost, when the balance hits $0
  • Active serverless worker discounts are not published, only offered through sales

Who Runpod is for

Runpod fits a solo developer or small ML team that wants a specific GPU today, pays by card and can tolerate some setup. Community Cloud is the cheapest fixed-price H100 in this group for experiments and notebooks. Secure Cloud is the column to use for anything customers depend on, given what Runpod's own docs say about Community reliability.

Look elsewhere in three cases. A team that wants Python functions turned into autoscaling GPU jobs with no containers to manage should price Modal, which also covers sandboxes in our AI agent sandboxes ranking. A team that only needs a CPU server or a small VPS will pay far less on the hosts in our cheap VPS guide or the Hetzner review. And anyone running a model locally should read the Ollama review before renting a GPU at all.

Best Runpod alternatives

If Runpod is not the right fit, these are the closest options.

ToolBest forStarts at
RunpodDevelopers who want a specific GPU by the second, from consumer cards to H200 and B300H100 PCIe Pod $1.99/hr Community or $2.89/hr SecureVisit →
LambdaTeams that want data-center GPUs from a single provider, from one card to large clusters1x H100 PCIe $3.29/GPU/hrVisit →
ModalPython teams that want serverless GPU functions with no containers to manageFree plan · Team $250/mo · custom-priced plans tooVisit →
ReplicateDevelopers who want to run or publish models behind an API with the least setupH100 $0.001525/sec ($5.49/hr)Visit →
Vast.aiCost-driven users who will pick hosts from a live marketplaceMarketplace: H100 PCIe from $2.53/hr (median $2.85)Visit →
Lambda
A data-center GPU cloud priced above the Runpod Secure column, with clusters from 16 GPUs.
Visit →
Modal
Serverless GPUs driven from Python code, billed per second, with a free monthly credit.
Visit →
Replicate
A model-hosting platform that is the most cross-shopped GPU cloud on Toolradar, at the highest H100 rate here.
Visit →
Vast.ai
A GPU marketplace where live listings can undercut every fixed price on this page.
Visit →

The bottom line

Runpod is worth it for hourly GPU work if you pick the right product: Community Pods for cheap experiments, Secure Pods for anything that must stay up, and flex serverless only for endpoints idle most of the day. Watch the two traps, parked volume disks and a zero balance on Pods without a network volume, and the bill stays predictable.

For the full hourly comparison, read the cheapest GPU cloud table; for serverless options beyond GPUs, the serverless platforms ranking and AWS Lambda pricing; for where to keep model weights, object storage pricing. Toolradar profiles: Runpod, Lambda, Modal and Replicate. Subscribe free for the Devshot brief.

Methodology: list prices and billing rules from each vendor's pricing page and docs, read on November 4, 2026; Vast.ai figures are live marketplace listings on that day. Totals are our arithmetic. Devshot ran no benchmarks, so nothing here reports speed, cold-start times or availability.

Cite this: Devshot, 'Runpod Review 2026: Pricing, Pros, Cons and Alternatives', November 2026.

Frequently asked questions

How much does Runpod cost in 2026?

Runpod bills per second. An H100 PCIe Pod is $1.99 an hour on Community Cloud or $2.89 on Secure Cloud, an A100 PCIe $1.19 or $1.79, and an RTX 4090 $0.34 or $0.89, per its pricing page updated October 28, 2026. Serverless H100 flex workers cost $5.39 an hour, and storage starts at $0.05 per GB a month.

What is the difference between Community Cloud and Secure Cloud?

Runpod's docs describe Community Cloud as peer-to-peer providers with variable reliability, suited to cost-sensitive work, and Secure Cloud as T3 and T4 data centers with high redundancy, suited to production and sensitive data. Secure costs more for every GPU in the price list.

Are Runpod serverless workers cheaper than pods?

Only for endpoints that sit idle most of the day. A flex H100 worker costs nearly twice the hourly rate of a Secure H100 Pod, but it scales to zero while the Pod bills whenever it runs. By our arithmetic, serverless wins while the endpoint is busy under about 13 hours a day.

Does Runpod charge for stopped pods?

Not for the GPU, but yes for storage. A stopped Pod's volume disk bills at $0.20 per GB a month, twice the running rate, and network volumes keep billing at $0.07 per GB under 1 TB. Container disks are erased when the Pod stops and are not charged.

Is Runpod cheaper than Lambda?

For a single H100 PCIe, yes: both Runpod columns undercut Lambda's $3.29 an hour. Lambda lists 8-GPU H100 nodes at $3.99 per GPU-hour and clusters of 16 to 256 H100s, which suit larger training jobs.

Does Runpod have a free tier?

No free tier appears on its pricing page; you deploy on prepaid credit, with at least one hour of credit needed for an on-demand Pod. For free GPU credit, Modal's Starter plan includes $30 of compute a month.
Related guides

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

Get the Devshot brief

Free daily newsletter, read in 5 minutes.

Subscribe free