Gemini Enterprise Pricing 2026: Dates Every Meter Turns On

On July 13, 2026, a billing switch flipped inside Google Cloud. Agent Gateway, the traffic-control layer that every “agentic” interaction on Google’s Gemini Enterprise Agent Platform passes through, started charging for usage. Almost nobody covered it, because Google never held a press event. The company simply printed the date on its pricing page, the way a landlord slides a rent notice under the door.

It was not the first switch, and it will not be the last. Skills Registry billing commenced on July 1. Semantic Governance Policy billing commences on August 1. Memory Bank and Sessions billing both commence on September 1. If your company spent the past nine months building Artificial Intelligence (AI) agents on Google’s platform while those services ran unmetered, this is the calendar on which the free ride ends.

The good news, and it is genuinely good news, is that the meters themselves are cheap. Run the arithmetic on Google’s published rates, as the worked example below does, and a busy 50-person deployment racks up about $47 a month in new meter charges, sitting next to $1,500 a month in seat subscriptions. The scary part of the bill is the boring part, and it was there all along.

Key Takeaways

• Four billing start dates are already printed on Google’s pricing page: July 1, July 13, August 1, and September 1, 2026, collectively switching on five platform meters.
• Per-seat subscriptions run from $21 per month, and Google’s small-business price now exactly matches Microsoft’s list price for its Copilot equivalent.
• A worked 50-person bill shows one line item swallowing about 97 percent of the total, and it is not the one the new billing dates point at.

What Just Started Billing on Google’s Agent Platform?

First, orient yourself, because Google has made the naming genuinely confusing. Gemini Enterprise is the umbrella product Google launched on October 9, 2025, billed in its own launch announcement as “the new front door for AI in the workplace.” At launch, CNBC put the prices at $30 per person monthly for large organizations and $21 for smaller ones, the same seat prices Google’s pages still show in mid-July 2026. It absorbed Agentspace, Google’s earlier agent product, whose URL now redirects to the Gemini Enterprise page. Underneath the app sits the Gemini Enterprise Agent Platform, the plumbing layer where companies build and run their own custom agents.

That plumbing layer is where the new meters live. Google prices it with three unified resource types: Agent Compute (billed per virtual CPU hour, or vCPU-h, the standard unit of rented processor time), Agent Memory (per gibibyte-hour of RAM), and Agent Storage (per gibibyte-month, a gibibyte being the binary cousin of a gigabyte). Services that are really “per request” get converted into compute hours at published exchange rates.

Here is the full schedule, straight from the pricing page as of mid-July 2026:

Notice the pattern. Google is not repricing anything. It is turning on billing for services that already existed and already had rates printed next to them, one service per month, through the end of summer. An Application Programming Interface (API) call your agent made through Agent Gateway in June cost nothing. The identical call on July 14 costs one fifteen-thousandth of $0.085.

How Much Does Gemini Enterprise Cost Per User?

The subscription side is simpler, though Google buries one detail worth knowing. The Business edition, aimed at small teams with “no IT setup needed,” starts at $21 per seat per month and includes 25 GiB of pooled storage and data indexing per seat. The Standard and Plus editions, aimed at large organizations that need enterprise-grade controls like customer-managed encryption keys and data residency, start at $30 per seat per month, with Plus pricing available through sales. A Frontline edition exists for deskless workers at organizations already running at least 150 Standard or Plus seats, with a slim 2 GiB storage allotment.

The buried detail: Google’s own pages currently disagree about the Business edition’s seat cap. The product page says 1 to 300 seats; the documentation’s edition-comparison table says 1 to 500. That kind of loose end tells you the packaging is still settling, which is worth remembering before you sign a 12-month commitment.

One warning for anyone searching old articles: this Gemini Enterprise is not the “Gemini for Workspace” add-on Google sold under a nearly identical name in 2024. Google stopped selling those add-ons on January 15, 2025, and folded Gemini directly into Workspace plans instead. The October 2025 product is a separate Google Cloud platform. If a pricing article predates October 2025, it describes a product that no longer exists.

Is Gemini Enterprise Free? What the Free Tiers Actually Cover

Not free, but padded. Every consumption resource carries a monthly free tier per account: the first 50 vCPU-hours of Agent Compute, the first 100 gibibyte-hours of Agent Memory, and the first 1 gibibyte-month of Agent Storage cost $0.00. The seat subscriptions offer 30-day trials rather than free tiers.

Fifty free compute hours is a meaningful buffer. At Agent Gateway’s exchange rate of 15,000 calls per vCPU-hour, the free tier alone absorbs 750,000 gateway calls a month before the first cent is billed, assuming your agents are not burning those hours on other compute. Small pilots will feel nothing in July. The meters matter at production scale, and that is exactly the point of turning them on now: October’s launch cohort has had nine months to reach production scale.

The Two Bills: Why Seats Are 97 Percent of the Money

Here is the arithmetic that reframes the whole story. Picture an illustrative 50-person company on Standard seats, running a reasonably busy custom agent: 3 million gateway calls a month, 100 GiB of accumulated skills, memories, and session state, plus a million memory writes and 3 million reads.

The consumption side, at published rates:

$$\frac{3{,}000{,}000 \text{ calls}}{15{,}000 \text{ calls/vCPU-h}} = 200 \text{ vCPU-h}$$

Two hundred vCPU-hours at $0.085 each comes to $17.00 for the month’s gateway traffic.

Storage adds 100 GiB at $0.30 each: $30.00. The operation charges are rounding errors: 3 million reads convert to a single vCPU-hour ($0.085), and 1 million writes to another. Total meter bill: about $47, before the free tier shaves it further.

The subscription side: 50 seats at $30 each is $1,500 per month.

So the complete bill lands near $1,547, and the seats are roughly 97 percent of it. All five of those carefully dated billing switches, the ones this article exists to explain, govern the other 3 percent. If you run agent workloads that are wildly heavier than this sketch, your mileage will differ, and token charges for the underlying models are always billed separately on top. But for most organizations the strategic read is clear: Google priced the meters to be ignorable and the seats to be the business. The meters exist so that the heaviest builders pay their way, and so that every workload is at least accounted for once the platform’s accounting is fully switched on.

Gemini Enterprise vs Microsoft Copilot Pricing

The seat prices only make sense next to the competition, and the small-business tier is where it gets pointed. Gemini Business: $21 per seat per month. Microsoft’s equivalent add-on, Microsoft 365 Copilot Business, lists at exactly $21.00 per user per month, though Microsoft is currently running it at a promotional $18.00 for annual commitments started between July 1 and September 30, 2026, first year only. A price match at the list line and a discount at the checkout line, in the same month Google’s meters started running, is what a seat-price war looks like in practice.

Upmarket, the shapes converge too. Gemini Enterprise Standard starts at $30 per seat. Licensing guides put Microsoft’s enterprise Copilot add-on at the same $30 per user per month it has carried since launch. Neither company is competing on the sticker. They are competing on what the sticker drags along: Google wants your agents living on its Cloud platform paying consumption rates, and Microsoft wants you locked deeper into the 365 bundle.

What Happens September 1: Memory Rent Arrives

The September 1 switches deserve their own attention, because they change an architectural assumption, not just a bill.

Memory Bank is the service that lets an agent remember things across conversations. Sessions is the service that stores the conversations themselves. From September 1, both are billed at $0.30 per GiB-month for everything stored, and Memory Bank’s storage charge explicitly includes revisions. In plain terms: an agent’s memory stops being a free byproduct and becomes rent. An agent that remembers everything forever is no longer a clever design choice; it is a small recurring liability that compounds every month, the way an unattended log bucket does.

Anyone who lived through the last cloud-cost crackdown knows what happens next. Companies audited idle virtual machines and orphaned storage buckets then; they will audit zombie agents and hoarded memories now. Financial operations (FinOps) teams, the people whose job is policing cloud spend, are about to inherit a new asset class. The practical homework before September 1 is boring and worth doing: inventory which agents accumulate memory and session state, set retention rules, and delete what no one will miss, because from that date forward, forgetting is a cost optimization.

The Google Maps Lesson Google Learned

If the phrase “Google platform, free while you build, then the meter arrives” gives you a chill, your memory is working. In 2018, Google repriced the Google Maps API with roughly ten weeks of warning, raising the cost of map loads from $0.50 to $7 per thousand, a 14-fold increase that blindsided thousands of small developers when it took effect on July 16, 2018.

The 2026 version is the photographic negative of that episode. The rates were published before the billing started. The dates were printed months out. The unit prices are measured in fractions of a cent, with free tiers underneath. Whatever else you think of the agent platform’s economics, the rollout suggests Google understood exactly how much trust the Maps repricing burned, and priced this rollout to avoid a sequel. The risk to watch is not the mid-2026 rates. It is what happens to them after the platform wins.

That story already has a broader arc on this site: June 2026 was the month the industry’s per-token prices kept falling while the bills kept growing, as GitHub moved Copilot to usage-based billing and agentic workloads multiplied consumption. Google flipping on its agent meters in July and September is the same tide reaching the next shore. And the agents doing the metered calling are increasingly wired into Google’s own data tools through the Model Context Protocol, the plumbing covered in Google’s push to make Maps and BigQuery agent-ready.

Mark the date: on September 1, every memory your agents have been quietly accumulating starts costing $0.30 per gibibyte, every month, until someone decides what they are allowed to forget.