Google now bundles Gemini into Workspace Business and Enterprise plans rather than selling the former $20 and $30 per-user add-ons separately, which raised Workspace list prices by about $2 per user per month, while Gemini Code Assist lists at $19 to $45 per user per month and the Vertex AI Gemini API is billed per token. The headline simplification hides a real cost shift: the AI capability that used to be an optional add-on is now baked into the base plan price, so every seat pays for it whether or not the person uses it.
The three ways Gemini is sold
Gemini reaches enterprise buyers through three distinct commercial doors, and conflating them is the most common pricing mistake. The first is Gemini inside Google Workspace, the assistant in Gmail, Docs, Sheets, and Meet, now bundled into the Workspace plan rather than sold as a separate seat. The second is Gemini Code Assist, the developer assistant priced per developer seat. The third is the Gemini model family on Vertex AI, consumed through the API and billed per token of input and output. Each has its own pricing logic, its own discount path, and its own right-sizing question, and a buyer negotiating one should not assume the terms carry to the others.
Gemini in Workspace: the bundling change
Until early 2025, Gemini in Workspace was an add-on: Gemini Business at roughly $20 per user per month and Gemini Enterprise at roughly $30, layered on top of the Workspace subscription. Google then folded the assistant into the core Workspace Business and Enterprise plans and retired the standalone add-ons, raising the base plan price by about $2 per user per month. The table sets out the before and after.
| Model | Before (2024) | After (2026) |
|---|---|---|
| Gemini Business add-on | ~$20 / user / month | Retired, bundled into base plan |
| Gemini Enterprise add-on | ~$30 / user / month | Retired, bundled into base plan |
| Workspace base plan | Add-on optional | ~$2 / user / month higher, AI included |
| Who pays for AI | Only seats that opted in | Every licensed seat |
For an organization where most users would have bought the add-on anyway, bundling is a price cut, since $2 is far below the old $20 to $30. For an organization where only a fraction of seats wanted AI, bundling is a price rise, because every seat now carries the cost. The right response depends on your real adoption, and the underlying Workspace plan ladder this sits on is covered in our Google Workspace enterprise pricing 2026 guide.
Gemini Code Assist pricing
Gemini Code Assist is priced per developer seat in two tiers, and unlike the Workspace assistant it remains a deliberate add-on you buy only for the developers who use it.
| Tier | List price | Built for |
|---|---|---|
| Code Assist Standard | ~$19 / user / month (annual) | Code completion and chat in the IDE |
| Code Assist Enterprise | ~$45 / user / month (annual) | Codebase customization, larger context, admin controls |
| Monthly (no annual commit) | Higher per-seat rate | Short pilots, flexibility over price |
Because Code Assist is a per-developer add-on, the right-sizing question is which engineers genuinely use it, not a blanket rollout. The competing developer assistant from Microsoft creates a pricing fight here, covered in our Microsoft Copilot licensing guide.
Vertex AI Gemini API token pricing
The Gemini models on Vertex AI are consumed through the API and billed per token, separately for input and output, with rates that vary by model size. Lighter, faster models cost a fraction of the largest, most capable models per token, and long context windows and multimodal inputs carry their own pricing. The practical implication is that API cost scales with usage, not seats, so the control is prompt and architecture design, choosing the smallest model that meets the quality bar, caching where the same query repeats, and keeping context windows tight. For a workload of any size, the per-token rate is negotiable inside a committed-spend agreement on Vertex AI, which folds into the broader Google commercial position at Google Cloud advisory.
How enterprise discounts apply
Each Gemini door discounts differently. The Workspace assistant, now bundled, discounts as part of the Workspace seat negotiation, so the lever is the overall Workspace deal rather than a separate AI line. Code Assist discounts on volume and term like any per-seat subscription. The Vertex API discounts through committed spend, where a forecast token volume buys a lower per-token rate. The buyer's job is to negotiate all three inside the single Google relationship rather than treating each as a standalone purchase, because consolidated commitment is what moves the discount. How the three assistants compare on price and capability is the subject of our Gemini vs Copilot vs ChatGPT enterprise guide, and the wider Gemini footprint sits in Gemini enterprise.
The bundled-AI lever: Bundling Gemini into Workspace removed the ability to decline AI on a seat, so the cost is now structural in the base plan. The lever that remains is seat count and seat tier. If only a portion of your workforce uses the assistant, the saving is not in declining AI, which you can no longer do, but in placing low-need users on a lower Workspace tier and reserving the higher tier for the population that genuinely uses the advanced capability. Audit real usage before renewing every seat at the top tier the bundle nudges you toward.
Right-sizing AI seats
The cost discipline for enterprise AI is the same one that governs every per-seat product: pay for the seats that produce value and stop paying for the ones that do not. With Gemini now bundled into Workspace, the question shifts from whether to add AI to which Workspace tier each user needs, since the assistant rides along with the tier. With Code Assist, the question is which developers actually use it, measured by real activity rather than by headcount. With the Vertex API, the question is whether the architecture uses the smallest adequate model and caches repeated work. Across all three, the saving comes from matching the spend to genuine use, not from a blanket decision in either direction.
Governing the AI spend
AI cost drifts faster than most software because adoption and usage both move. Review Workspace seat tiers at renewal against real activity, review Code Assist seats against developer usage data, and monitor Vertex API token consumption monthly because a single new feature can multiply token volume overnight. Put all three on the same renewal and review calendar as the rest of the Google relationship so they are negotiated together and the committed-spend math stays current. Buyers who set AI seats once and forget them pay for adoption that never materialized; buyers who review usage capture the saving the bundle and the per-token model both reward.
Data handling and the enterprise tier premium
Part of what separates the higher Gemini tiers from the lower ones is data handling, and for many enterprises that, not raw capability, is what justifies the price. The enterprise tiers carry commitments on how prompts and outputs are handled, whether data is used to train models, and what administrative and compliance controls are available, which are the controls a regulated organization requires before putting business data into an assistant. A buyer evaluating the tier choice should separate the capability premium from the governance premium, because a team that needs the data-handling commitments has a different justification for the higher tier than a team that merely wants better output. Pricing the governance requirement explicitly prevents both overpaying for capability nobody uses and underbuying the controls compliance demands.
Measuring real adoption before you scale
The cost discipline that matters most for an AI assistant is measuring adoption before scaling the seat count, because the gap between licensed seats and active users is wider for AI tools than for almost any other software category. A pilot that measures who actually uses the assistant, how often, and for what produces the data that sizes the rollout, and that data routinely shows that a fraction of the licensed population drives most of the value. Scaling to the measured active population rather than to headcount is what keeps an AI deployment from becoming the most expensive shelfware in the estate. The same right-sizing discipline applies across every per-seat product, and the broader approach sits in our Google Workspace enterprise pricing 2026 guide.
Using the competitive field as a lever
The enterprise AI assistant market has more than one credible option, and that competition is the buyer's strongest pricing lever. Google's Gemini, Microsoft's Copilot, and the standalone enterprise offerings each compete for the same seats, and a buyer who can genuinely consider more than one prices better than one locked to a single vendor by default. The work is to keep the alternatives credible, with a real evaluation rather than a token one, so the incumbent vendor prices against the risk of losing the seats rather than against a list it knows the buyer will accept. How the three compare on price and capability is the subject of our Gemini vs Copilot vs ChatGPT enterprise guide.
Controlling Vertex API cost at the architecture level
Because the Vertex AI Gemini API bills per token rather than per seat, its cost is an engineering problem as much as a procurement one, and the largest savings come from architecture rather than negotiation. Choosing the smallest model that meets the quality bar for each task, rather than defaulting every call to the most capable model, can cut per-token cost by a wide margin, since lighter models cost a fraction of the largest ones. Caching responses where the same query recurs avoids paying twice for identical work. Keeping prompts and context windows tight reduces the token count on every call. And batching or rate-limiting non-urgent work prevents runaway consumption from a single misbehaving process. These controls are set in the application, not the contract, which is why API cost discipline belongs to the engineering team with finance visibility, and why a token-billed service needs monitoring the way no per-seat product does. The committed-spend agreement on top of disciplined usage is what produces the lowest effective rate, and that agreement folds into the wider position at Google Cloud advisory.
The buyer's takeaway
Gemini pricing in 2026 is three products wearing one name, and the buyer who prices them separately wins. Treat the Workspace bundle as a base-plan tier decision, buy Code Assist only for developers who use it, and control Vertex API cost through model choice and committed spend. Negotiate all three inside the single Google relationship, and review real usage before renewing seats at the top tier the bundle nudges you toward. We model Gemini cost and negotiate the AI terms through our cloud contract negotiation and SaaS license optimization practices. The cheapest AI seat is the one a person actually uses; the rest is shelfware with a smarter name.