Oracle put a meter on Fusion AI. AI Units cost one cent each, 20,000 come free every month, and the model you pick, not the agent, drives the bill. This pillar prices the 26C model and maps the levers to set before you upgrade.
Oracle has named the meter for Fusion AI. In the 26C release, AI Units are the single currency for every agentic action, at one cent each and pooled into one balance. Every Fusion Cloud customer gets 20,000 free each month. This pillar prices the model and sets the buyer side moves before your upgrade.
Oracle AI Units cost one cent each and are the single consumption currency for Fusion AI from the 26C release forward. Oracle replaced its earlier token based and agent type based pricing with one outcome aligned meter. Every AI action, from answering a question to generating a document, draws units from one pooled balance shared across every Fusion pillar.
The unit is a measure of value, not a raw token count. Oracle defines what each action consumes in its Fusion AI Unit Actions Table, and the rate varies by the kind of work and the model behind it.
You can read the mechanics on Oracle's own AI Agents for Fusion Applications page and the pooled unit definitions in the Oracle Fusion Cloud global price list. Read both before you accept a sizing from your account team.
One balance covers agentic work everywhere in the suite. That breadth is deliberate. Oracle wants one meter for every agent you run.
For a buyer, the unified pool is a double edged thing. It simplifies the invoice to one meter. It also concentrates every AI workload onto a single currency whose rate Oracle controls. Detail sits in the AI Units explained companion guide.
An AI action is any metered task an agent or embedded feature performs, and Oracle rates each one in its Fusion AI Unit Actions Table. The point is that not all actions cost the same. A simple lookup is cheap, a long document generation is not, and a multi step agent run stacks several actions together.
Because a single agent run can chain many actions, the meter rewards efficient design. An agent that retrieves once and writes once costs far less than one that loops. Design and model choice, together, set the real unit cost of a workflow.
Every Oracle Fusion Cloud customer receives 20,000 AI Units per month at no charge under the 26C model. That included allowance is enough to cover light, routine usage across the suite before you buy a single incremental unit. It is the reason many customers will see no AI line at all in the first months after upgrade.
Consumption above the monthly allowance is billed, and additional capacity is sold in packs. The packs behave differently from the free grant, and the difference matters for how you buy.
The allowance is per customer, not per user or per pillar, so it is one shared grant across your whole Fusion estate. A large enterprise and a mid market customer receive the same 20,000 units. For smaller estates that grant can cover most routine agentic work indefinitely. For large estates it is a starting credit that heavier usage will exceed.
Treat the free grant as the floor of your forecast, not the ceiling of your usage. Measure how fast real agentic adoption consumes it, then plan packs around the overage. The teams that track burn weekly in the first quarter after upgrade avoid both surprise overage and premature overbuying.
The Oracle AI Unit purchase model at a glance
| Layer | What it is | Price | What to watch |
|---|---|---|---|
| Free allowance | Included AI Units every Fusion Cloud customer gets | 20,000 per month, free | Use it or lose it, resets monthly |
| Incremental pack | Pooled AI Units bought beyond the allowance | About $1,000 per 100,000 units | Carries over in the service period, size to proven volume |
| Basic LLM actions | General actions on the GPT oss based tier | $0.00 in AI Units | Pin routine work here to protect the allowance |
Purchased packs carry over during the service period rather than resetting each month. That removes the pressure to consume on a monthly clock. It also means a pack sized to a peak month sits as unused capacity if your real burn is lower. Buy against demonstrated consumption, not a projected spike.
AI Agent Studio is included with your Oracle Fusion SaaS Cloud subscription at no additional license cost. It ships with a library of ready to use agent templates and the tooling to configure, extend, and orchestrate agents against your Fusion data.
Oracle introduced the studio as a retention play, and the studio itself is free. You can read the launch detail on the Oracle Fusion Insider blog. The cost is never the tool. It is the consumption the agents drive.
What is free is the studio and the templates. What is metered is the consumption those agents drive, denominated in AI Units, and what is separately licensed is production scale custom agent work. Keeping those three ideas distinct is the whole budgeting exercise.
The full inventory, and the line between the free studio and the chargeable pieces, sits in the what AI Agent Studio includes guide.
You need a Custom AI Agent subscription when you move beyond configuring delivered agents into building and running custom agents at production scale. Oracle documents the boundary in its Custom AI Agent subscription documentation. The studio is free to use, but the entitlement to publish and operate custom agents is a separate subscription.
This is the seam where a proof of concept quietly becomes a licensed deployment. A team can prototype in the studio for free, then cross into subscription territory the moment the custom agent goes live. Knowing where that line sits protects your budget and your negotiation.
Keep prototyping and production separate, and know exactly which environment your custom agents run in. The subscription line is crossed by operating a custom agent at production scale, so an agent that graduates from a sandbox to a live business process is the event to watch. Governance around promotion, not the studio itself, is what protects the budget.
The distinction sounds procedural. It is commercial. A single unreviewed promotion can convert a free experiment into a licensed deployment, so the promotion gate is the cheapest control you can put in place before your 26C upgrade goes live.
The studio is free, the templates are free, and the first 20,000 units a month are free. The cost is the model you pick and the moment you publish to production.
White Paper · Oracle
The Oracle Fusion AI Agents Playbook
The AI Unit cost model, the LLM trigger, the Agentic Applications fee, and the 26C levers. Read it free.
The large language model you choose drives the AI Unit bill, not the type of agent. This is the single most important fact in the 26C model. General actions on the Basic LLM tier, based on the GPT oss model, are priced at zero AI Units. Premium frontier models consume units at a higher rate.
Two teams can run an identical agent and produce very different invoices purely on model selection. The lever is not whether you use AI. It is which model does the routine work.
How model choice moves the AI Unit cost of the same work
| Model tier | Typical use | Relative AI Unit cost | Buyer move |
|---|---|---|---|
| Basic (GPT oss) | General actions, routine drafting, lookups | $0.00 for general actions | Default routine work here to protect the free allowance |
| Standard | Balanced quality and cost for common agents | Metered per action | Use where Basic quality falls short |
| Premium frontier | Complex reasoning, high value analysis | Highest per action | Reserve for work that justifies the rate |
Treat model choice as a spend control, not a technical default. The team that sets model policy before rollout keeps the meter predictable.
Source: Redress Compliance advisory engagement file, 2024 to 2025, against Oracle published pricing.
The Fusion Agentic Applications subscription is a separate platform fee, distinct from AI Unit consumption, that unlocks the right to publish agentic applications to production and access the Agentic Applications Builder. Oracle introduced it alongside 26C, announced on 24 March 2026 in the Fusion Agentic Applications announcement and the Agentic Applications Builder expansion.
The platform fee bundles a large annual AI Unit allowance on top of the base 20,000 monthly units, along with the builder and the production publishing right. The trigger is important. It is publishing an agentic application to a production environment, not upgrading to 26C and not enabling a feature flag.
The standard account team framing is that the Agentic Applications platform fee and a large upfront AI Unit commitment are a necessary part of moving to 26C. We disagree. In the Fusion engagements we advised on, the platform fee is triggered only by publishing an agentic application to production, and forecast based unit commitments routinely overshot proven consumption by 20 to 40 percent. Most customers get real value from delivered agents and the free allowance long before any platform fee applies. The buyer side move is to separate what 26C includes for free from what production publishing actually triggers, model your burn bottom up, and commit to the platform fee and unit volume only when your own usage, not an Oracle projection, justifies it.
The full mechanics sit in the Agentic Applications subscription guide.
The strongest lever is timing, because 26C is when AI consumption starts to bill. Organizations testing agentic features in non production ahead of upgrade are not yet accruing charges, which gives you a window to model real usage before money moves. Use it.
Everything else follows from measuring before you commit. The levers below are the ones we run on Fusion AI deals.
These moves sit inside the wider Fusion renewal covered on the Oracle practice page and the Fusion SaaS renewal playbook.
Model policy moves the most money, because it is the only lever that changes the unit cost of work you have already decided to do. The others control how much you buy. Model policy controls what each purchase is worth. A default set to the Basic tier can stretch the free allowance across a surprising share of routine agentic work.
Run all three together. Set the model default low, measure real burn in the non production window, then take a single, informed AI package into the Fusion renewal. That sequence, in that order, is how the prepared buyers in our engagement file kept Fusion AI predictable.
Walk a representative upgrade. The numbers are illustrative, but the moves are the ones we run. A 4,000 employee enterprise is upgrading to 26C and its Oracle account team proposes a large annual AI Unit commitment plus the Agentic Applications platform fee, sized to an Oracle adoption forecast.
The pitch bundles a multi million unit annual pool and the platform fee into the renewal as a single AI package. It is framed as the price of being ready for agentic Fusion. Both halves contain a real point and a real trap.
The response is not to reject AI Units. It is to separate proven from projected, and to fold the meter into the existing agreement.
Vendor proposal versus buyer side counter
| Lever | Vendor proposal | Buyer side counter |
|---|---|---|
| Unit volume | Large pool on an adoption forecast | Sized to bottom up burn by pillar and model |
| Model policy | Premium models by default | Basic tier default, premium gated by value |
| Platform fee | Signed with the upgrade | Signed only when production publishing is real |
| Rate protection | Discount at signing only | Unit rate and pack discount held for the term |
The outcome is the capability without the overcommitment. The buyer keeps the free allowance working, defaults routine actions to the zero cost tier, defers the platform fee until production is real, and buys units against proof. The pool it declined was never sized to its actual use.
Oracle AI Units are part of a broad industry shift from per seat software to per use metering, and the same discipline applies across all of it. SAP meters Joule on AI Units, Microsoft meters Copilot on credits, and Workday meters agents on Flex Credits. The mechanics differ, the buyer posture does not.
Oracle's model has two features that reward a prepared buyer. The free 20,000 unit monthly allowance is more generous than most rivals give, and purchased packs carry over rather than expiring monthly. Set against that, the platform fee for agentic applications and the model driven cost curve are where an unprepared buyer overspends.
The cross vendor view sits alongside the Microsoft Copilot Credits analysis. Treat every usage meter as a line to model, cap, and benchmark, never as a flat fee to wave through.
Use this sequence. It works whether you are planning your 26C upgrade or already live on it.
Oracle AI Units cost one cent each and are the single consumption currency for Fusion AI in the 26C release. Every metered AI action, from answering a question to generating a document, draws AI Units from one pooled balance across every Fusion pillar. Each Fusion Cloud customer gets 20,000 AI Units free per month before any charge applies.
Every Oracle Fusion Cloud customer receives 20,000 AI Units per month at no charge under the 26C model. That included allowance covers light usage across HCM, ERP, SCM, and CX before you buy anything. Consumption beyond the monthly allowance is billed, and additional capacity is sold in 100,000 unit packs that carry over rather than expiring at month end.
Yes. AI Agent Studio is included with your Oracle Fusion SaaS Cloud subscription at no additional license cost, and ships with ready to use agent templates. What is not free is the consumption those agents drive, metered in AI Units, and the separate Custom AI Agent subscription that governs building and running custom agents at production scale.
The chosen large language model drives the AI Unit bill, not the type of agent. General actions on the Basic LLM tier, based on GPT oss, are priced at zero AI Units, while premium frontier models consume units at a higher rate. Two teams running the same agent can produce very different bills purely on model selection.
The Fusion Agentic Applications subscription is a separate platform fee, distinct from AI Unit consumption, that unlocks the right to publish agentic applications to production and access the Agentic Applications Builder. Oracle's platform fee SKU bundles a large annual AI Unit allowance on top of the base 20,000 monthly units. The trigger is publishing to production, not upgrading to 26C.
Release 26C, landing around July 2026, is when Oracle's AI consumption model starts to bite commercially. Organizations testing agentic features in non production environments ahead of their 26C upgrade are not yet accumulating billable AI Unit consumption. The SKUs for pooled AI Units became orderable on the May 2026 global price list.
Purchased AI Unit packs carry over during the service period rather than resetting monthly, which is a meaningful difference from meters that expire unused capacity every month. The 20,000 unit monthly free allowance is use it or lose it. Buy incremental packs against demonstrated consumption so committed capacity is not stranded.
Model your real AI Unit consumption bottom up by pillar and by model, then buy against proven volume rather than an Oracle forecast. Pin the Basic tier for routine actions, reserve premium models for high value work, and negotiate the platform fee and unit rate together inside your Fusion renewal.
The AI Unit cost model, the free allowance math, the LLM cost trigger, the Agentic Applications platform fee, and the 26C levers to set before you upgrade.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders sizing Fusion AI before the next renewal.
500+ enterprise clients. 11 vendor practices. We will model your Fusion AI Unit burn, set your model policy, and fold the platform fee and unit rate into your renewal before Oracle sizes them for you.
One email a month on Oracle AI Units, Fusion renewals, ULA certification, and audit defense. Buyer side only.
