RunPod review: pricing, G2 and Trustpilot, September 2026

RunPod merits a GPU-compute shortlist, with separate storage costs and capacity checks kept in the buying decision.¹ G2 and Trustpilot report different rating aggregates; neither supplies a workload-specific reliability measurement.^2,3 The published prices and rating aggregates were checked 28 September 2026.^1,2,3 No first-hand use.

TL;DR

• Choose the compute model before comparing prices: RunPod publishes separate rates for Pods, Serverless and Clusters.¹ A Pod price is insufficient for a Serverless budget.
• Keep retained storage in the budget. RunPod lists a higher Volume Disk rate for idle storage than for running storage.¹
• The published ordinary Cluster limit is up to 64 GPUs.¹ RunPod markets Reserved Clusters separately for enterprises scaling to 10,000+ GPUs, with pricing through sales.¹
• Treat the ratings as a reason to investigate fit. Both review sites’ AI-generated summaries mention GPU availability concerns, without measuring a deployment failure rate.^2,3

What does RunPod sell?

RunPod sells Pods for dedicated GPU instances, Serverless for API inference and Clusters for multi-node jobs.¹ The product choice changes both the published tariff and the workload it addresses.¹

Compute model	RunPod’s published purpose	Example from the hourly price view
Pods	Dedicated instances for development and long-running jobs	H100 PCIe: $2.89/hour1
Serverless	Inference workers billed on usage	H100: $4.79/hour1
Clusters	Multi-node workloads	H200 SXM: $4.31/hour displayed in the GPU row1

The distinction belongs at the start of a compute infrastructure evaluation. Compare the service needed for the job before treating two hourly figures as alternatives.

RunPod describes workload duration, traffic pattern and control needs as inputs to that choice.¹ For a development environment, start with the Pod configuration. For an inference endpoint, assess the Serverless tariff and worker configuration. A distributed job needs the Cluster terms and a quote for the complete deployment.

What GPU prices does RunPod publish?

RunPod’s checked Pod price view lists the RTX A5000 at $0.27/hour, H100 PCIe at $2.89/hour and B200 at $6.79/hour.¹ Its pricing page reports an update on 27 September 2026.¹ These are published rates, not a quote for confirmed capacity.

The selected configurations below retain RunPod’s displayed prices and resource quantities, checked 28 September 2026.¹ The Pod section shows both Community Cloud and Secure Cloud labels, but the captured view does not resolve which mode supplies each price.¹ Confirm the cloud mode, currency and tax treatment before using an amount in a purchase budget.

Pod GPU	Published GPU memory	Host RAM	vCPUs	Published $/hour
RTX A5000	24 GB VRAM	25 GB	9	$0.271
RTX 4090	24 GB VRAM	41 GB	6	$0.741
RTX 5090	32 GB VRAM	35 GB	9	$0.991
RTX A6000	48 GB VRAM	50 GB	9	$0.531
RTX 6000 Ada	48 GB VRAM	167 GB	10	$0.841
RTX Pro 6000	96 GB VRAM	188 GB	16	$2.091
A100 PCIe	80 GB VRAM	117 GB	8	$1.591
A100 SXM	80 GB VRAM	125 GB	16	$1.591
H100 PCIe	80 GB VRAM	188 GB	16	$2.891
H100 SXM	80 GB VRAM	125 GB	20	$3.491
H100 NVL	94 GB VRAM	94 GB	16	$3.191
H200	141 GB VRAM	276 GB	24	$4.591
B200	180 GB VRAM	283 GB	28	$6.791
B300	288 GB HBM3e	251 GB	32	$7.891

“H100” alone is too broad a label for a price comparison. RunPod lists PCIe, SXM and NVL variants with different prices and configurations.¹ The A100 PCIe and SXM rows share a displayed price and GPU memory, while their host RAM and vCPU allocations differ.¹ Keep the full configuration attached to the rate.

Serverless has a separate hourly view: A100 at $2.72, H100 at $4.79, H200 at $5.93 and B200 at $8.64.¹ Its 4090 entry is $1.10/hour.¹ The capture does not resolve an active-versus-flex worker tariff, so these figures should stay labelled as the displayed hourly view.¹ Use the RunPod pricing analysis for the pricing-specific follow-up, then confirm the selected tariff in the deployment quote.

What costs sit outside the GPU rate?

RunPod publishes storage charges separately, including Volume Disk at $0.10/GB/month while running and $0.20/GB/month while idle.¹ An idle storage tariff means the GPU rate alone cannot describe the cost of keeping an environment between jobs.¹

Storage type or state	Published $/GB/month
Container Disk	$0.101
Volume Disk, running	$0.101
Volume Disk, idle	$0.201
Network Storage, Standard, under 1 TB	$0.071
Network Storage, Standard, over 1 TB	$0.051
Network Storage, High-Performance	$0.141

The standard network-storage labels distinguish “under 1TB” from “over 1TB”.¹ Confirm the rule at exactly that boundary and how the tier applies before estimating a large volume’s bill.

Build a budget around the chosen compute tariff and the storage that remains allocated. Record running storage and idle storage separately. RunPod itself identifies storage and deployment choices as total-cost factors.¹ A comparison based only on the GPU-hour figure leaves those published charges out.¹

What limits does RunPod publish?

RunPod’s ordinary Clusters section advertises scaling up to 64 GPUs; its separate Reserved Clusters section targets enterprises scaling to 10,000+ GPUs.¹ The reserved statement should not be read as the ordinary Cluster limit or as capacity already available to every account.

For ordinary Clusters, RunPod describes no commitments and shared storage.¹ The price view displays H200 SXM at $4.31/hour and A100 SXM at $1.79/hour, while L40S, H100 SXM and B200 say “Contact sales”.¹ Those GPU-row prices should not be treated as a complete multi-node job quote.

RunPod markets Reserved Clusters with guaranteed availability, custom configurations and SLA-backed uptime.¹ Each displayed reserved GPU entry requires sales contact.¹ Require the actual capacity commitment and SLA terms if either controls the decision; the marketing phrase supplies no uptime percentage.¹

The Pod catalogue also states configuration limits through its GPU memory, host RAM and vCPU quantities.¹ Treat those as published configurations to match against a workload, without assuming they establish throughput or suitability for a particular model.

What do G2 and Trustpilot report?

RunPod’s G2 listing displayed 4.6/5 from 49 reviews, while Trustpilot displayed a TrustScore of 3.8 from 315 reviews, checked 28 September 2026.^2,3 Keep the aggregates separate: they have different review populations and are published by different services.^2,3

Published field, checked 28 September 2026	G2	Trustpilot
Rating aggregate	4.6/52	TrustScore 3.83
Review count	492	3153
Five-star share	83%2	64%3
One-star share	2%2	19%3

Trustpilot’s larger displayed review population also has a higher one-star share.^2,3 That establishes a difference between the published review distributions. It cannot explain the cause of the difference or supply an uptime estimate.

Review collection needs context. Visible G2 reviews include “Incentivized” and “Source: Seller invite” labels.² Trustpilot’s profile says RunPod invites customers to review.³ These records do not establish what share of all G2 reviews was incentivized, or why the two aggregates differ.

Both sites publish AI-generated review summaries that mention positive pricing themes and GPU availability concerns.^2,3 Trustpilot’s summary says it considers 151 reviews, a different population from its 315-review aggregate.³ Use those themes to choose evaluation questions; they are not independently measured incidents.

Under the research methodology, retain the checked date and the collecting service with each aggregate. Keep the review population separate from the workload whose performance needs testing.

What should change the buying decision?

Glamdring Research’s verdict is to evaluate RunPod against the exact GPU configuration, compute model and retained-storage budget needed for the job. The published catalogue supports that shortlist, while the review aggregates leave workload-specific capacity and reliability unresolved.^1,2,3

For an evaluation, compare the cloud GPU providers on a matching workload. Record capacity in the required region, time until the environment is ready, successful job completion and the resulting bill.

The decision changes when that evaluation demonstrates acceptable completion cost and repeatable access. If a guaranteed capacity commitment or uptime term is essential, obtain the reserved contract RunPod advertises before treating either as secured.¹

Frequently asked questions

Does RunPod list the same B300 memory for Pods and Serverless?

No. RunPod’s checked pricing page lists 288 GB HBM3e for a B300 Pod and 280 GB in the Serverless B300 entry.¹ Confirm the deployable configuration for the chosen product; the two published quantities should not be substituted for each other.

What reservation durations does RunPod display?

RunPod’s Reserved Clusters price section displays 1mo, 3mo, 6mo, 12mo and 12mo+ options.¹ The displayed GPU entries still say “Contact sales”.¹ A duration option therefore needs a quote before it supplies a budget.

Does a verified Trustpilot review verify RunPod’s performance?

No. Trustpilot says it may label a review “Verified” when it can confirm a business interaction, while it does not verify specific reviewer claims.³ Read the label as an interaction check, not a benchmark result.