Google Gemini Enterprise

Google Gemini Enterprise on cost: seats for the app, a meter for the platform

Gemini Enterprise is two purchases that buyers sometimes treat as one. The Gemini Enterprise app is billed per seat per month¹. Underneath, the Agent Platform is charged by consumption², and Google sells the two as separate purchases³. A finance team at a 1,000 employee company therefore needs two forecasts: how many people will hold seats, and how much agent and model consumption those people and their agents will generate.

Google's documentation combines three roles in one product, covering intranet search, assistant and agent platform⁴; the FAQ adds that the app brings Google AI to every employee⁵. The Agent Platform grew out of Vertex AI, in Google's account⁶. This profile places Gemini Enterprise third in the catalog, and the reasons are mostly about which prices Google publishes and which it does not.

The editions and what is priced

Several editions of the app exist, each cut for a different kind of organization⁷. Business is listed from $21 a seat each month⁸, and Google meant it for small firms and teams with at most 300 people⁹. The two larger editions lift the seat cap entirely, raise quotas, include Gemini Code Assist and let users reach agents built by other companies¹⁰. On top of either, a Frontline edition for frontline staff can be bought as an extra^11,12.

Only the Business price is public. What a Standard or Plus seat costs is not publicly stated on the pages we opened; Google's pricing URL returned a not found page¹³. For a 1,000 employee company that is the central fact of this profile, because Business stops at 300 and the editions that fit a larger company require a quote.

Working the Business numbers

The Business price still helps as a benchmark. On our arithmetic, at the starting price of $21:

1. 100 seats cost $2,100 a month, or $25,200 a year⁸. 2. 300 seats, the edition's ceiling, cost $6,300 a month, or $75,600 a year^8,9. 3. 1,000 seats at the same rate would cost $21,000 a month, or $252,000 a year, though Business does not cover that size⁹.

The last line is a reference point only. It shows the scale of a seat budget at 1,000 people if Standard or Plus were priced near Business, which the pages we opened do not say.

Terms are monthly or yearly¹⁴, and the Cloud console limits how many seats one subscription can hold¹⁵. Each subscription belongs to one billing account¹⁶; each license belongs to one project in one location¹⁷. Licenses are allocated from the subscription to projects and locations¹⁸ and can be assigned automatically or manually¹⁹. License assignment comes after role assignment: each user must hold Gemini Enterprise User at minimum²⁰. For finance, those rules mean that seat counts are managed per project, which can help charge costs back to departments.

The consumption side

On a metered subscription there is no license fee²¹; each month's bill reflects that month's consumption²². Usage of partner models is metered too²³. Self serve Cloud billing accounts can now buy prepaid credits²⁴. Usage rates per query or per token are not publicly stated on the pages we opened.

The combination gives a buyer flexibility. A company can hold seats for heavy users and run lighter or automated workloads on consumption. It also means that the total cost depends on a usage forecast that only a pilot can provide.

Quotas: the ceiling inside each seat

A seat buys a quota. Per license and per day, Standard permits 160 Assistant queries and 5 images²⁵, and Plus permits 200 queries²⁶. Feature quotas are shared among an edition's users, storage and indexing excepted²⁷; those two are shared across every edition in the same project and location²⁸. Admins can turn on a limit for any single user to restrict consumption per user²⁹.

Pooling is favorable for a 1,000 employee company with uneven usage: light users leave quota for heavy users. The single user limit gives finance a way to stop one team from consuming the pool.

Dates that affect renewal

Existing Standard and Plus subscriptions lose their bundled Code Assist entitlement on March 31, 2027³⁰. Any business case that counted Code Assist inside the seat value needs a renewal model that either drops it or prices it on its own.

Notebook Enterprise can be subscribed for in quantities as large as 5,000 licenses³¹. Google pairs that research and knowledge assistant with Gemini Enterprise³².

Trials

Both Standard and Plus come with 30 free days³³; Plus can even be trialed through the landing page without speaking to sales³⁴. Business edition can also be tried at no cost for 30 days³⁵, with an account created from a work email³⁶. A month is enough to measure query volumes against the quotas above.

How the product works

Google's documentation mentions ready connectors to popular third party apps³⁷ and a search that respects permissions on company information³⁸. Business has a separate help center; the main documentation serves the other editions³⁹. Agent Gateway is the door for agent traffic in both directions⁴⁰, and tool use policies are written as plain sentences⁴¹.

Google's example prompts are ordinary office work, such as summarizing the best performing campaigns from last year⁴². Business edition examples cover stock tracking and purchase orders⁴³, customer emails informed by Google Search⁴⁴ and onboarding with Gemini Notebook⁴⁵. Those examples suggest a seat population of everyday knowledge workers, which is the population a seat price grows with.

Implementation effort

Google's pages describe setup as licensing and administration more than as a service engagement. Staff can build agents themselves in a no code workbench⁴⁶. A Deep Research agent comes prebuilt⁴⁷; Data Insights Agent is still a preview, limited to Standard and Plus⁴⁸. Custom agents made with ADK or the Agent Platform publish directly into the app for Standard and Plus customers⁴⁹; agents from elsewhere join through the A2A protocol⁵⁰. An Agent Marketplace offers agents made by partners⁵¹.

A packaged implementation service with its own price is not publicly stated on the pages we opened, and neither is a standard implementation timeline. For a 1,000 employee rollout, the buyer should budget internal Google Cloud administration time and, where custom agents are needed, developer time on the Agent Platform.

Integrations

ServiceNow, SharePoint, Jira and Confluence have connectors out of the box⁵². Workspace data from Gmail, Drive, Calendar and Groups syncs in real time⁵³. The Business edition reaches Notion, HubSpot, Workspace and Microsoft 365⁵⁴. BigQuery can be queried through a federated connector⁵⁵, which leaves the data where it already sits⁵⁶. Connectors to third party apps call public endpoints beyond Google's own network⁵⁷, a detail for the security review.

Security and compliance

Google's compliance page, which applies to Standard, Plus and Notebook Enterprise⁵⁸, runs through FedRAMP, PCI DSS, HIPAA, ISO 27001 and SOC 1 to 3 among others⁵⁹. That breadth is useful to procurement.

Customers may hold their own encryption keys⁶⁰, managed in Cloud KMS on Standard and Plus⁶¹ or in an external key manager⁶². Choosing the global region rules out both CMEK and Access Transparency⁶³. Data can be kept in the us or eu location, or placed in global⁶⁴. Google says it does not train its foundation models on prompts, answers or connected company data⁶⁵, and that each customer's data is kept apart inside its own project⁶⁶. Being SaaS, it does not accept custom network controls set by the customer⁶⁷. Deletion requests are completed within 60 days⁶⁸.

Data handling details for the security review

Enterprise editions offer Cloud Audit Logs and Access Transparency⁶⁹. Google's editions table states that, for enterprise customers, prompts, responses and data do not train foundation models, which sets them apart from consumer Gemini⁷⁰. Business runs as a web service at business.gemini.google⁷¹, with the same promise that customer inputs and outputs will not train Google's models⁷². Standard and Plus offer residency in a region the customer selects⁷³.

On the Agent Platform, training or fine tuning on customer data needs the customer's prior permission or instruction⁷⁴. For Models as a Service, Google holds customer data only for limited periods⁷⁵. Request and response logs are off unless enabled⁷⁶, and Search grounding stores logs no longer than three days⁷⁷.

Sensitive Data Protection can block sensitive records from being indexed via connectors⁷⁸. It works in all editions, Business included⁷⁹, and in three multi regions⁸⁰. Its content policies skip the older type of data store, which includes BigQuery tables, Cloud Storage buckets and website sources⁸¹; encrypted Microsoft files are not checked either⁸². Google advises Workforce Identity Federation for connectors to outside apps⁸³. A company that runs VPC Service Controls must name every permitted Data Cloud source in an organization policy⁸⁴; Google blocks new data stores for any source left off that list⁸⁵. Each of these is a configuration task for the buyer's cloud team, and it belongs in the implementation estimate.

Questions to put to Google

1. What is the per seat price of Standard and Plus at 1,000 seats, monthly and annual^13,14? 2. What consumption rates apply on the Agent Platform for the agents we plan²? 3. How will renewal pricing change once the Code Assist entitlement ends³⁰? 4. Which implementation services are available, and at what price?

Governance

Google says the app gives one central view for governing every team's agents⁸⁶. Enterprise administrators work through IAM, federated workforce identity for SSO, and the ACLs of each connected system⁸⁷. Document level permissions apply⁸⁸. Model Armor comes in all editions⁸⁹ and checks prompts and tool responses live to stop injection⁹⁰. Every agent gets its own identity⁹¹, protected from the start by Context Aware Access⁹².

Business admins manage team membership and decide who may use which connector⁹³, without any project level IAM to administer⁹⁴.

Where it fits on cost

The natural buyer already runs Google Cloud and Workspace, wants a seat priced assistant for office staff across functions such as finance, HR and sales⁹⁵, and can hold a separate consumption budget for agents. Teams of up to 300 can price it from the public page^8,9. Government and education institutions are offered tailored terms⁹⁶.

At 1,000 employees, the price of the editions that fit is not publicly stated¹³, and the two purchases must be forecast separately³.

Why it ranks third

Gemini Enterprise places third because Google publishes a real seat price for Business and a clear consumption model for the Agent Platform^8,2,21, backed by a broad certification list⁵⁹. It ranks below Copilot Studio because the editions a 1,000 employee company needs are unpriced on the pages we opened¹³, while Copilot Studio's pack price applies at any size⁹⁷. Nexus sits above it for two reasons: its price is set per role automated and ignores headcount, and its consultancy support carries no per diem or consultant time^98,99, while Gemini's app cost rises with seats¹ and a packaged implementation service is not publicly stated on the pages we opened.