CAST AI vs Kubecost: Pricing, Kubecost vs OpenCost and Which Kubernetes Cost Tool to Buy
CAST AI vs Kubecost is the wrong question in about half the cases where it gets asked, because the two products do different jobs and the vendors say so themselves. Kubecost measures: it allocates Kubernetes spend to namespaces, deployments, labels and teams and produces rightsizing recommendations that a person applies. CAST AI acts: it replaces the cluster autoscaler, picks and buys nodes, moves workloads onto spot capacity and resizes pod requests without a ticket. The real decision is whether you are buying a report or a controller, how much of each is free, and what the paid tiers cost, which neither vendor states on its own pricing page but both state on AWS Marketplace. This page puts those two rate cards side by side for the first time, explains the 250-core line that decides whether Kubecost costs you anything, and covers OpenCost, ScaleOps and StormForge for the teams whose shortlist is really wider than two.
projected this month if unattended
Spend by team
Budget forecast
The short answer
CAST AI and Kubecost are complementary more often than competing. Kubecost is a cost allocation and reporting tool with a free tier up to 250 cores and a paid Enterprise edition listed on AWS Marketplace at $15,000 a year for up to 250 cores; CAST AI is a cluster automation platform listed on AWS Marketplace at $1,000 a month plus $5.00 per managed CPU per month, with a free monitoring plan. Buy Kubecost when you need to show teams what they spend, buy CAST AI when you want the spend cut automatically, and run both when you need the second measured independently of the first.
Costanalyst appears in the table below, so here is the plain version first. We are neither an in-cluster cost agent nor an automation controller. We read your AWS, Azure and Google Cloud bills and your Kubernetes cost data read-only, allocate them per team and product, and publish our prices. We sit next to either of these tools as the independent measurement, not in place of them.
How we compared
Five things that actually separate these tools
Report or controller: what each product actually does
Settle this before you look at a single price, because it decides whether you are comparing or combining. Kubecost is built on the open source OpenCost allocation engine and does one job very well: it turns Kubernetes resource usage and your cloud bill into cost per namespace, deployment, label, pod and team, reconciled against what the cloud provider actually invoiced, and it surfaces rightsizing and idle-resource recommendations. It does not apply them. CAST AI's own comparison of the two products states that plainly: Kubecost does not execute changes, it surfaces rightsizing recommendations that engineers must implement manually. CAST AI, by contrast, installs a controller that takes over node provisioning, bin-packs pods onto fewer nodes, moves eligible workloads to spot capacity with fallback to on-demand, live-migrates containers, and resizes pod CPU and memory requests. Kubecost tells you the cluster is 40 percent idle; CAST AI removes the idle nodes. If your problem is that nobody can see who spends what, you need the first. If your problem is that everyone can see it and nobody has time to act, you need the second. If you have both problems, which is most platform teams over about 500 cores, you are reading a comparison of two products you will probably end up running together.
What each vendor publishes on price, and where
Neither vendor puts a number on its own pricing page, and both put full rate cards on AWS Marketplace, sold under their own seller names. CAST AI's pricing page is a form headed "Get a custom quote" that says its pricing model depends on factors specific to your environment. Its EKS listing on AWS Marketplace (prodview-vtvxyzbzs3huy, seller Cast AI) prices a Free plan at $0 for unlimited Kubernetes monitoring and cost reduction insights, Growth at $1,000 a month or $12,000 a year for up to four managed clusters and 500 CPUs, GrowthPro at the same $1,000 a month for unlimited clusters and 2,000 CPUs, Enterprise at $5,000 a month or $60,000 a year, and a metered charge of $0.00694444 per managed CPU per hour, which is $5.00 per CPU for a 720-hour month, with the note "We do not currently offer refunds." Kubecost's pricing page, kubecost.com/pricing, now redirects to IBM's Apptio site, which lists Foundations as always free with unlimited clusters up to 250 cores and 15-day metric retention, and Enterprise as contact-sales. But IBM Software sells two public Kubecost offers on AWS Marketplace: IBM Kubecost Enterprise (prodview-zdyeh4lqc5jaw) at $15,000 for twelve months, with the listing stating "The public offer provides a 12-month subscription for up to 250 cores. For all other packaging options, please request a private offer", and IBM Kubecost Cloud (prodview-qnvpdqk2wrsui), the single-tenant SaaS edition, at $7,950 for twelve months plus an overage dimension of $33.40 per core. Both IBM listings disallow quantity changes and state that refunds, cancellations and credits are issued on a case-by-case basis. Every figure on this page was read off those listings and vendor pages on 17 September 2026.
The 250-core line, and what $15,000 actually buys
Read the two Kubecost numbers together and something odd falls out: the free Foundations tier covers up to 250 cores, and the paid Enterprise public offer also covers up to 250 cores. The $15,000 is not buying scale. It buys the Enterprise feature set on the same estate: a unified multi-cluster view, unlimited metric retention instead of 15 days, role-based access control, enterprise integrations, enhanced GPU optimization, resource quota automations and dedicated support. Above 250 cores, the price is a private offer, and IBM's product page adds the line "For deployments above 3,200 VPCs, please contact us." So the budget question for Kubecost is not "how many cores do we run" but "do we need RBAC, long retention and a single pane across clusters." A team with three clusters and 200 cores that only wants a chargeback report can stay on Foundations indefinitely. A team with the same 200 cores that needs finance to log in with restricted views and pull twelve months of history is on Enterprise at $15,000 a year, which works out to $5.00 per core per month at the cap. The third-party figures you will find in search results, $8 per node per month and a Business tier at about $3.42 per vCPU, are not on any IBM page or listing we read; the $3.42 appears in a competitor's blog and on a per-container-hour basis in another, and should not go in a budget request.
What CAST AI costs at your size, and the savings-share confusion
CAST AI's Marketplace shape is a platform fee plus a meter. At 250 managed CPUs on Growth, that is $12,000 a year plus 250 times $5.00 times twelve, or $15,000, for $27,000 a year. At 500 CPUs, the Growth cap, it is $42,000. At 2,000 CPUs on GrowthPro, $12,000 plus $120,000, or $132,000. At 5,000 CPUs on Enterprise, $60,000 plus $300,000, or $360,000. Those are list prices from a public offer; enterprise deals close as private offers below them, but the public card is the ceiling you negotiate down from and the only first-party number in existence. Here is the part that trips buyers up. CAST AI's own blog comparing itself to Kubecost describes a savings-share model of roughly 15 to 20 percent of generated savings, and works an example where a $20,000 a month cluster bill cut in half produces a fee of $1,500 to $2,000 a month. That model is not on the Marketplace listing, which prices flat plans plus a per-CPU meter, and it is not on cast.ai/pricing, which prices nothing. Treat the two as different commercial motions that the same vendor runs, ask which one your quote is on, and do the arithmetic for both, because at high savings rates the per-CPU meter is cheaper and at low savings rates the share is.
What you have to let it touch
Kubecost is read-mostly. It runs in the cluster, needs Prometheus (its own or yours), reads metrics and your billing export, and writes nothing to workloads. Its only production risk is the resource footprint of the in-cluster components, which competitors describe as heavy at scale, and which is the reason the SaaS edition exists. CAST AI starts read-only and then asks for a great deal more. Its onboarding installs a lightweight read-only agent via a single script that produces a savings estimate; enabling automation then hands the controller the right to create and delete nodes, evict and reschedule pods, change instance types, and move workloads onto spot capacity. That is exactly what you are paying for and exactly what a security review will ask about. Two things follow. First, CAST AI cannot run alongside your existing Karpenter or Cluster Autoscaler on the same cluster; it replaces them. Second, the vendor that acts on the cluster is also the vendor that reports the savings, so the number in its dashboard is a vendor-measured number. Kubecost, OpenCost or a read-only bill-level tool is how you check it.
At a glance
6 Kubernetes cost tools compared
← Scroll to see all columns →
| Tool | Best for | What it does to the cluster | Access it needs | Pricing |
|---|---|---|---|---|
| Kubecost (IBM) | Cost allocation, showback and chargeback per namespace and team; free to 250 cores | Measures and recommends; applies nothing | In-cluster agent plus Prometheus, read-only to workloads | Foundations free to 250 cores; Enterprise $15,000 per year to 250 cores on AWS Marketplace, private offer above; Cloud edition $7,950 per year plus $33.40 per core overage |
| CAST AI | Estates that want nodes, spot capacity and pod requests managed automatically | Replaces the autoscaler; creates and removes nodes, moves pods, resizes requests | Read-only agent first, then write access to nodes and workloads | Free monitoring plan; Growth $1,000 per month plus $5.00 per managed CPU per month on AWS Marketplace; Enterprise $5,000 per month; no refunds |
| OpenCost | Teams that want free, open source cost allocation and will run it themselves | Measures; the allocation engine Kubecost is built on | In-cluster, reads Prometheus metrics | Free, open source, CNCF project |
| ScaleOps | Fully autonomous pods, nodes, spot and GPUs, including air-gapped | Rightsizes pods and nodes, places pods, manages GPUs, autonomous from install | Self-hosted controller with write access; in-cluster Prometheus per cluster | $50,000 per year platform fee plus $9.00 per vCPU on AWS Marketplace; metering period not stated |
| StormForge Optimize Live (CloudBolt) | Pod request rightsizing only, priced per vCPU with no minimum | Resizes pod CPU and memory requests in place; nodes and spot are out of scope | Three in-cluster pods; only usage metrics leave the cluster | $0.0041 per requested vCPU-hour pay-as-you-go, or $32,400 per year per 1,000 vCPUs on AWS Marketplace; 30-day trial |
| Costanalyst | An independent, read-only Kubernetes cost baseline per team, next to either tool | Nothing; it never touches the cluster | Read-only billing and Kubernetes cost data | $99, $299 and $799 per month, published; Kubernetes on the Scale plan |
Product facts checked July 2026. Vendors change pricing and packaging often, so confirm before you buy.
Tool by tool
What each tool is genuinely best at
Kubecost (IBM)
Best for: Cost allocation, showback and chargeback per namespace and team; free to 250 cores
Kubecost is the default answer to "what does each team spend on Kubernetes," and since IBM acquired it in September 2024 it has been pulled into the Apptio FinOps stack, which is why its pricing page now lands on apptio.com. The free tier is real and generous: unlimited clusters up to 250 cores, 15-day retention, unlimited users, community support. Enterprise adds the multi-cluster view, unlimited retention, RBAC and support, and its public Marketplace offer is $15,000 a year for the same 250-core ceiling. Its weakness is the one CAST AI keeps pointing at: it recommends and does not act, so the savings it identifies only happen if an engineer has time to make them happen.
Kubecost (IBM) compared to CostanalystCAST AI
Best for: Estates that want nodes, spot capacity and pod requests managed automatically
CAST AI is the automation platform, and its published capability list points entirely at the cluster: autoscaler with bin-packing, spot automation with on-demand fallback, container live migration, commitment utilization, pod mutations, a rebalancer, memory event handling and vertical plus horizontal pod autoscaling, across EKS, GKE, AKS and Oracle Cloud. Its site publishes no price; its Marketplace listing publishes a full one. The published customer results (Akamai 40 to 70 percent, Yotpo 40 percent) and the LeanOps benchmark it cites (45 to 58 percent compute reduction) are vendor-reported and should be checked against your own bill during the trial.
CAST AI compared to CostanalystOpenCost
Best for: Teams that want free, open source cost allocation and will run it themselves
OpenCost is the open source, CNCF-hosted allocation engine that Kubecost's free tier is built on, and for a single cluster where you want cost per namespace and label with no vendor relationship, it is enough. What it does not give you is the multi-cluster unified view, the cloud-bill reconciliation depth, the saved reports, RBAC and support that separate Kubecost Enterprise from Foundations, and it has no automation at all. If you are choosing between Kubecost Foundations and OpenCost, the honest difference is packaging and a UI; if you are choosing between OpenCost and CAST AI, you are choosing between a free report and a paid controller.
ScaleOps
Best for: Fully autonomous pods, nodes, spot and GPUs, including air-gapped
ScaleOps competes with CAST AI, not with Kubecost, and lands on this page because buyers who reject Kubecost for not acting usually shortlist both automation vendors. Its scope is wider than CAST AI on GPUs and air-gapped deployment, and its Marketplace card is a large fixed fee that favors big estates. The ScaleOps vs StormForge page carries the full comparison of the automation vendors on how they apply a change to a running pod.
ScaleOps compared to CostanalystStormForge Optimize Live (CloudBolt)
Best for: Pod request rightsizing only, priced per vCPU with no minimum
StormForge is the narrow, cheap automation option. It does the one thing Kubecost recommends and does not do, applies it in place without restarting pods, and stops there. For a team that already has Kubecost for allocation and only wants the rightsizing recommendations acted on, StormForge plus Kubecost is a cheaper stack than CAST AI, at the cost of leaving node provisioning and spot to Karpenter or the cloud provider.
StormForge Optimize Live (CloudBolt) compared to CostanalystCostanalyst
Best for: An independent, read-only Kubernetes cost baseline per team, next to either tool
We belong on this page for one reason. Whichever automation vendor you buy will report its own savings, and whichever allocation tool you buy will live inside the cluster the automation vendor changes. A bill-level, read-only view of Kubernetes cost per team and product, taken before the trial and read after it, is how finance confirms the number the vendor invoices against. That is the whole job, and the prices are on the page.
See how Costanalyst worksHow to choose
Pick by the problem you actually have
You need chargeback or showback per team and nobody has it today
Start with Kubecost Foundations, which is free up to 250 cores, or OpenCost if you would rather run the engine yourself. Allocation is the prerequisite for everything else on this page: an automation vendor's savings claim means nothing until you can say which team's spend it reduced. Move to Enterprise only when you hit the retention, RBAC or multi-cluster limits.
You already have the reports and the recommendations are not getting applied
That is the CAST AI case, and it is the most common reason a Kubecost customer ends up buying CAST AI rather than replacing it. Price it off the Marketplace card at your CPU count before the call, decide whether you want the whole platform (Growth or GrowthPro) or only the rightsizing layer (StormForge is cheaper for that), and run the read-only agent for a full weekly cycle before you switch automation on.
You run under 250 cores and have a small budget
Kubecost Foundations costs nothing and covers you. CAST AI's free plan covers monitoring but any automation puts you on Growth at $12,000 a year plus the meter, which at 200 CPUs is about $24,000. If you want automation at that size, StormForge pay-as-you-go at roughly $2.99 per vCPU-month with no minimum is the cheapest paid option, and the Kubernetes VPA is free.
You need RBAC, twelve months of history or one view across many clusters
That is Kubecost Enterprise, $15,000 a year on the public offer to 250 cores or a private offer above it. Ask IBM for the per-core price above 250 in writing; the public listing does not carry it. Kubecost Cloud at $7,950 a year plus $33.40 per core overage is the option if you do not want to operate the in-cluster components yourself, but confirm what core count the entitlement covers, because the listing does not say.
Security will not approve a controller with write access to nodes this quarter
Then CAST AI and ScaleOps are off the table until it does, and the comparison collapses to Kubecost, OpenCost and a bill-level tool. CAST AI's read-only agent can still produce a savings estimate for the business case without changing anything; use it to size the prize and take the write-access request to security with a number attached.
You want the savings measured by someone who is not being paid on them
Run both a measurement tool and the automation vendor, and keep the measurement outside the cluster the vendor changes. Kubecost or OpenCost inside the cluster plus a read-only bill-level baseline from Costanalyst is the arrangement that lets finance sign off the number. CAST AI's own comparison says most mature teams run both an optimization layer and a cost attribution layer; take the vendor at its word on that one.
Questions buyers ask
Kubernetes cost tools, answered
CAST AI vs Kubecost: which one do I need?
You need Kubecost if the problem is visibility: who spends what on Kubernetes, per namespace, team and label, with a free tier up to 250 cores. You need CAST AI if the problem is action: the waste is known and nobody has time to remove it, and you are willing to let a controller manage nodes, spot capacity and pod sizes. Teams over a few hundred cores commonly run both, with Kubecost or OpenCost measuring and CAST AI acting.
Is CAST AI better than Kubecost?
It is not better at Kubecost's job. Kubecost allocates cost more granularly and reconciles it against the cloud bill; CAST AI's cost reporting exists to justify its automation. CAST AI is better at reducing spend without engineering time, which Kubecost does not attempt. Compare them on the job you are hiring for, not on a single feature grid.
How much does Kubecost cost?
Kubecost Foundations is free with unlimited clusters up to 250 cores and 15-day metric retention. IBM Kubecost Enterprise is listed on AWS Marketplace at $15,000 for twelve months for up to 250 cores, with larger deployments sold as private offers. IBM Kubecost Cloud, the SaaS edition, is listed at $7,950 for twelve months plus $33.40 per core overage. The vendor's own pricing page redirects to apptio.com and shows no dollar figure.
Is Kubecost free?
Yes, up to 250 cores. The Foundations tier is described by IBM as always free, with unlimited clusters up to 250 cores, 15-day metric retention, unlimited users and community support. Above 250 cores, or for RBAC, unlimited retention and a unified multi-cluster view, you need Enterprise, which is $15,000 a year on the public AWS Marketplace offer.
How much does CAST AI cost per CPU?
CAST AI's AWS Marketplace listing meters $0.00694444 per managed CPU per hour, which is $5.00 per CPU for a 720-hour month, on top of a platform fee of $1,000 a month for Growth or GrowthPro and $5,000 a month for Enterprise. A 500-CPU cluster on Growth therefore lists at about $42,000 a year. Its own pricing page publishes no figure and says pricing depends on your environment.
Kubecost vs OpenCost: what is the difference?
OpenCost is the open source, CNCF-hosted cost allocation engine; Kubecost is the commercial product built on it. Kubecost Foundations adds a packaged UI, cloud-bill reconciliation and saved reports for free up to 250 cores; Kubecost Enterprise adds the multi-cluster view, unlimited retention, RBAC and support for a fee. Neither one automates changes.
Can I use CAST AI and Kubecost together?
Yes, and CAST AI recommends it: its own comparison says most mature teams run both, with CAST AI as the optimization layer and Kubecost for cost attribution and chargeback. The two do not conflict because Kubecost writes nothing to workloads. Keep an eye on the resource footprint of running both sets of in-cluster components on small nodes.
Does Kubecost automate rightsizing?
No. Kubecost produces rightsizing and idle-resource recommendations and leaves the change to an engineer, which CAST AI's comparison states directly. If you want the recommendation applied automatically, that is CAST AI, ScaleOps or StormForge, or the open source Vertical Pod Autoscaler with the caveat that its Auto mode is deprecated in VPA 1.4.0.
Is Kubecost open source?
Partly. The allocation engine underneath it, OpenCost, is open source under the CNCF. Kubecost itself is a commercial product with a free Foundations tier and paid Enterprise editions, owned by IBM since September 2024. If you need a fully open source deployment, run OpenCost directly.
Does CAST AI replace Karpenter or the Cluster Autoscaler?
Yes. CAST AI takes over node provisioning and cannot run alongside Karpenter or the Cluster Autoscaler on the same cluster. If your team runs Karpenter well, benchmark node utilization, spot share and pending-pod time before you replace it; CAST AI's read-only agent will estimate the gap without changing anything.
Why does CAST AI's blog describe a savings-share price when its Marketplace listing shows flat plans?
Because the vendor runs more than one commercial model. The Marketplace listing prices Growth at $1,000 a month plus $5.00 per CPU and Enterprise at $5,000 a month; the vendor's blog describes a share of roughly 15 to 20 percent of savings. Ask which model your quote is on and price both against your expected savings rate, because they cross over.
Keep reading
See your savings in dollars
Connect your cloud and SaaS spend read-only and get a prioritized savings plan. Money never moves. Create your account and get started.