Copilot vs Gemini vs Amazon Q in 2026. What each costs, and how to choose.

Microsoft 365 Copilot and Gemini Enterprise list at $30 per user per month, and Amazon's assistant at $20. Suite fit, data exposure and real seat use matter more than that gap, and all three vendors discount committed deals.

How do Copilot, Gemini and Amazon Q compare on price?

At list, Microsoft 365 Copilot costs $30 per user per month, Google's Gemini sits at the same $30 and Amazon Q Business Pro costs $20. Every first proposal opens at those figures, and committed enterprise deals close below them.

Two of the three products have been repackaged since launch, so the same figure can now buy a different product. Before you compare quotes, confirm which SKU the account team is pricing.

What each list price covers in 2026

• Microsoft 365 Copilot. $30 per user per month paid yearly, or $31.50 billed monthly on an annual commitment, on top of a qualifying Microsoft 365 plan. Copilot Chat, without grounding in your work data, comes at no extra cost with eligible subscriptions. Copilot Business, for Microsoft 365 Business plans, lists at $21 and currently shows a reduced $18. Our Copilot pricing guide covers the SKUs in detail.
• Google Gemini. Google withdrew its separate Gemini Business and Gemini Enterprise Workspace add ons in January 2025 and folded Gemini into the price of Workspace Business and Enterprise plans. The $30 figure now belongs to Gemini Enterprise, Google Cloud's agent platform: Standard and Plus editions start at $30 per seat, the Business edition at $21 for up to 300 seats, and none of them requires Workspace. See what Gemini in Workspace includes.
• Amazon Q Business. Pro is $20 per user per month and Lite is $3, with index capacity billed by the hour on top. AWS closed Q Business to new customers on July 31, 2026. It now points new buyers to Amazon Quick, whose Professional tier is $20 per user plus a $250 monthly infrastructure fee per account, with Enterprise at $40.

Which AI assistant should an enterprise choose?

For most enterprises, the right choice is the assistant built into the productivity suite your people already work in. An assistant earns its fee from the context it can read: your documents, mail, meetings and permissions. One attached to a foreign suite starts every comparison behind, whatever its model scores.

In practice, most organizations end up negotiating price and scope with their incumbent. The other two assistants still matter, because a priced alternative is what keeps the incumbent's proposal honest.

Why we would not pick the assistant on a model bake off

The usual advice is to run all three assistants on the same prompts and buy the one that writes best. We think that test answers the wrong question. Model quality between these vendors changes every few months, while the suite that holds your mail and files stays for years.

What decides value in the deals we see is how much of the workday the assistant reaches without a connector project, and how little it interrupts the way people already work. Run the pilot inside your main suite, measure use, and keep the challenger quoted in writing so its price is on the table when you negotiate.

When a second assistant makes sense

• Split suites. Organizations running Microsoft 365 and Workspace side by side, often after an acquisition, can license each assistant for its own population.
• AWS centric teams. Engineering and data groups whose knowledge sits in AWS hosted repositories may get more from Amazon's assistant, indexed through its connectors, than from a suite assistant.
• A Google challenger inside a Microsoft shop. Gemini Enterprise is sold without a Workspace requirement and connects to Microsoft 365 and SharePoint, which makes it a real competing quote even where Workspace is absent. Our Copilot versus Gemini Enterprise comparison covers that case.

What share of AI assistant seats do employees actually use?

Far fewer seats get used than most business cases assume. In our seat negotiations, weekly active use settled at 25 to 45 percent of licensed seats after the novelty quarter, with a median near 38 percent. An enterprise wide commitment at that rate means paying list for a majority of seats that sit idle.

The first quarter flatters every pilot. Usage spikes, screenshots circulate, and the enterprise wide proposal follows. Then weekly use settles, and many licensed users turn out to be people who opened the assistant twice.

Sizing the Copilot attach, and why the same logic applies to all three

Microsoft proposals tend to default to Copilot for 100 percent of users. The persona approach in our Microsoft negotiation guide right sizes the attach to 35 to 65 percent of active users, on usage evidence. Gemini and Amazon seats need the same test, applied before the commitment is signed.

How to measure weekly active use in each console

• Microsoft 365 admin center. Go to Reports, Usage, then Microsoft Copilot and the Copilot report. It shows enabled users, active users, the active users rate and each user's last activity date, over 7, 28, 90 or 180 days.
• Google Admin console. Go to Menu, Generative AI, Gemini reports. The org level view shows active Gemini users as a share of eligible users. The user level view shows active days in the past 28 days and a usage level from high to zero.
• Amazon Q Business console. The analytics dashboard reports weekly active users, monthly active users and total users for each application environment. Amazon Quick has its own analytics dashboard.

Pull weekly figures for 8 to 12 weeks after the launch month and export them before the negotiation starts. Launch month usage is inflated by curiosity, so leave it out of the baseline you negotiate from.

How much does a staged ramp save over licensing a whole division?

In our negotiations, buyers who staged seats against measured weekly use paid 30 to 50 percent less in year one than buyers who licensed whole divisions upfront. Staging gave up nothing, because the option to add seats stays open for the whole term.

A staged ramp combines a first cohort sized to demonstrated demand, expansion gates tied to measured weekly active thresholds, and a price for the expansion seats fixed at signature.

A worked example: 4,000 users on Copilot

Say a 4,000 person division is quoted Copilot at $30 per user per month, and pilot data suggest about 1,520 of them, close to the median active rate we see, will use it weekly. The table compares licensing everyone with a ramp of 2,000 seats for six months and 2,600 seats for the next six.

The ramp saves $612,000, or 42.5 percent, and the first cohort still carries 480 seats above measured demand. The discount row matters as much. Licensing everyone at 25 percent off still costs $252,000 more than the ramp at full list.

Expansion gates that hold up in a contract

• Threshold. The next tranche opens when weekly active use in the current cohort passes an agreed share, measured from the vendor's own admin report.
• Timing. Added seats co term with the original subscription, so the whole population renews on one date.
• Reallocation. You can move a license from an inactive user to a new one during the term. Our reallocation rights clause gives sample wording.

Seats are always available to add and almost never available to remove, so the first commitment is the one to size on evidence.

How much discount can you get on Copilot, Gemini and Amazon Q?

Committed multi year seat deals we advised on closed 20 to 30 percent below list at all three vendors, although every first proposal came in at list. The terms that moved price were the same at each table: staged ramps with locked expansion pricing, caps on the renewal uplift, and the consumption lines.

Price Copilot Credits against your Azure commitment, as our credits brief explains, and the Vertex AI side of Gemini as set out in our Gemini licensing guide. Settle both as separate lines before you size the seat commitment.

What a second suite is worth

Organizations running both Microsoft 365 and Workspace carried 8 to 15 extra discount points into assistant negotiations. A multi year exclusivity clause asks you to give that up, and the vendor sets the price of that clause in its own favor. Where your second suite is credible, refuse exclusivity, because those points are available again at every renewal.

What the account team will say, and what to say back

• "This is list price for everyone." Reply that comparable committed deals at all three vendors closed below list, and that you will commit when the price reflects that.
• "Licensing the whole department is the only way to drive adoption." Reply that adoption shows up in your own usage report, and that you will expand when the first cohort passes the agreed gate, at today's price.
• "Exclusivity gets you our best price." Ask for the price with and without the clause. The difference is what exclusivity costs you each year.
• "Q Business is closed to new customers, so move to Quick now." Price Quick Professional and Enterprise against your measured users, including the $250 per account fee, before you agree any migration date.

Contract terms to ask for

• Price hold. Expansion seats at the signature price for the whole term. See our price hold clause language.
• Uplift cap. A fixed maximum increase on the per user price at renewal, so the first year discount does not disappear in year two. Our uplift cap redline shows the wording.
• Reduction right. The right to cut seats at each anniversary to measured weekly use, even if limited to a set share of the total.
• Repackaging protection. If the vendor folds the assistant into a base plan, as Google did with Workspace in January 2025, you stop paying for it separately. See the bundling clause.
• Separate consumption lines. Copilot Credits, Amazon Quick agent hours and Vertex AI usage each priced on their own line, with their own cap.

What have we seen in GenAI seat negotiations from 2024 to 2026?

Across roughly 18 to 25 enterprise GenAI seat negotiations I advised between 2024 and 2026, the gap between licensed seats and weekly active users shaped the price more than anything else. Three patterns held across vendors and industries.

• The median settled near 38 percent. That was weekly active use against licensed seats once the novelty quarter had passed, whichever assistant was bought.
• Staged buyers lost nothing. They kept every option to expand, and the ones who expanded did so at prices locked at signature.
• Data exposure arrived before productivity did. Each assistant surfaces whatever its suite's permissions allow, so permission sprawl became answer sprawl on the first day.

Why a permission audit sets your seat ceiling

A permission audit before rollout is a security project with a licensing consequence. It defines the population that can safely use the assistant, and that population is the realistic ceiling on any seat commitment. What to check depends on the suite.

• Microsoft 365. Sites shared with everyone in the tenant, broken inheritance and stale guest access. Once at least one Copilot license is assigned, SharePoint Advanced Management adds site access reviews and restricted content discovery at no extra charge.
• Google Workspace. Files shared with anyone who has the link or with the whole domain, found through the Drive log events in the Admin console.
• Amazon Q Business and Amazon Quick. Whether each data source connector syncs access control lists, because answers follow the permissions the index ingests.

How does the choice change for a 500 user company and a 20,000 user enterprise?

At 500 users, none of the small business tiers fit, and there is less room for custom terms. Copilot Business attaches to Microsoft 365 Business plans, which stop at 300 users. Gemini Enterprise Business and Amazon Quick Plus also cap at 300.

A 500 user company therefore pays enterprise list prices with little discount room, so staging is where most of its savings come from. A 200 seat Copilot pilot costs $72,000 a year at list, against $180,000 for all 500 users.

What to do next

1. Pick the primary by suite fit. Keep the other two quoted in writing as the price check.
2. Measure weekly actives past the novelty quarter. Export the console reports before any commitment.
3. Audit permissions before rollout. Fix oversharing first, then treat the population that can safely hold a seat as the ceiling for the commitment.
4. Stage the ramp. Size the first cohort to demonstrated demand, tie expansion to active thresholds, and lock expansion pricing at signature.
5. Refuse exclusivity where you run two suites. Keep the discount points a second suite brings.
6. Split out the consumption lines. Negotiate credits, agent hours and cloud AI usage separately, each with a cap, before sizing seats.
7. Get support if you need it. Our GenAI practice runs the comparison and the vendor talks with you.

Frequently asked questions