Here is a fun experiment. Open your company’s AI dashboard and sort users by how often they used it last month.
You will usually find three groups. A handful of people who live in it, drafting, coding and summarizing all day. A big middle who use it a few times a week. And a long tail who logged in once, asked it to write a birthday message, and never came back.
Under the most common pricing model, all three groups cost exactly the same.
Now put a number on it. Imagine a 40-person company on a $20-a-seat annual plan. That is $9,600 a year. If 15 of those people open the tool once a month, roughly $3,600 of that bill buys almost nothing. Nobody decided to waste that money. The pricing model decided for them.
That is the seat tax. And it is only one of five ways AI tools charge for work in 2026.
Why the pricing model matters more than the price
Two AI tools can both advertise “$20” and cost wildly different amounts for the same team.
One charges per head. One charges per word processed. One hides the AI inside an office suite you already pay for. Another charges a small flat fee and lets you pay the model makers directly.
The headline price tells you what one unit costs. The pricing model tells you what the units are. And the units decide your bill far more than the number on the pricing page.
This matters more now than it did a year ago. AI use inside companies has stopped being an experiment. Teams are running AI agents, analyzing long documents and writing code with it every day. Heavier use means the gap between a good-fit pricing model and a bad-fit one can easily run into thousands of dollars a year, even for a small company.
So, let’s walk through all five models you are likely to meet, with real 2026 prices and worked examples for real team sizes.
Model 1: The flat seat
How it works: One price per person, per month, no matter how much or how little they use it.
Who uses it: This is the default for the big AI labs. ChatGPT Business and Claude Team both charge $25 per user per month, or $20 when billed yearly. OpenAI actually cut its standard Business seat by $5 in 2026, a sign that competition on seat prices is real.
The math for a 10-person team: $250 a month on monthly billing, or $2,400 a year on the annual plan. Simple, predictable, easy to approve.
Great for: Teams where nearly everyone uses AI daily. Budgeting is simple, finance teams love it, and nobody has to think about usage.
Watch out for: You pay for the person who never logs in. Ten unused seats on a $20 annual plan is $2,400 a year spent on nothing. Seat plans also come with fair-use limits, so your heaviest users can still hit a wall and get pushed toward a more expensive tier.
Model 2: The two-speed seat
How it works: A cheaper seat for most people, a much more expensive one for power users, mixed inside the same workspace.
Who uses it: In 2026 both OpenAI and Anthropic added Premium seats at $125 a month, or $100 when billed yearly, giving roughly five times the usage of a Standard seat. The reason is not hard to guess: agent-style AI, which plans and carries out multi-step tasks, burns through far more computing power than a simple chat.
The math for a 30-person team: Put 25 people on Standard and 5 on Premium, billed yearly. That is (25 × $20) + (5 × $100) = $1,000 a month, or $12,000 a year. Give everyone Premium “just in case” and the bill jumps to $3,000 a month. The mix saves $24,000 a year.
Great for: Mixed teams with a few developers, analysts or content leads running heavy workloads, and a majority with lighter needs.
Watch out for: Someone has to decide who gets the expensive seat, and revisit that list every quarter. Premium seats have a way of spreading, because nobody likes being told they are on the slower lane.
![A split illustration: a calm 20 km/h street with bicycles on the left, and a fast 100 km/h open road with a car on the right] (https://cdn.sanity.io/images/2343n0v8/production/9cd2f5ce2496c05fdcd1911e0a57e2ce50b33b5c-1599x1066.jpg?w=1000)
Model 3: The suite bundle
How it works: The AI comes inside office software you already pay for, either built in or sold as an add-on to it.
Who uses it: Google stopped selling Gemini separately and folded it into Workspace plans, so Business Standard at $14 per user per month (annual) or $16.80 (flexible) includes Gemini in Gmail, Docs, Sheets and Meet. Microsoft takes the add-on route: Microsoft 365 Copilot Business lists at $21 per user per month, with promotional pricing as low as $18 at points in 2026, on top of a Microsoft 365 plan such as Business Standard, which rose to $14 in July 2026.
The math for a 30-person team: On Google, about $420 a month on annual billing, and that covers email, storage and video calls too. On Microsoft, Copilot plus Business Standard comes to about $35 per person, or roughly $1,050 a month.
Great for: Companies whose whole day happens in Gmail and Docs, or Outlook and Excel. The AI works right where the files already are, with no new tool to learn.
Watch out for: You are tied to one vendor’s AI models. And the AI cost gets buried inside a bigger bill, which makes it hard to tell whether the AI part is actually paying off.
Model 4: Pure pay-as-you-go
How it works: No seats at all. You pay per token, which roughly means per word the AI reads and writes, through a provider’s API.
Who uses it: Every major AI lab sells API access, and larger enterprise contracts are moving in this direction too: more and more deals now charge a seat fee just for access, with actual usage billed separately at API rates.
The math: Hard to predict, and that is the point. A quick email summary might cost a fraction of a cent. Reviewing a 200-page contract costs many times more. Your bill tracks real work, not headcount.
Great for: Fairness. Light users cost little, heavy users cost more, and nobody pays for empty chairs. It is also the natural choice if you are building AI into your own product.
Watch out for: Raw API access is a developer tool. Without a proper chat interface, admin controls and spending limits on top, most of your team cannot use it, and your finance team cannot forecast it.
Model 5: Flat platform fee plus bring-your-own-keys
How it works: A hybrid of models 1 and 4. You pay a small fixed fee for the workspace itself: the chat interface, user management, security and admin controls. Then you plug in your own API keys and pay the AI providers directly for what your team actually uses.
Who uses it: This is where newer workspace tools like Grengin sit. Its pricing is tiered by team size rather than by seat: $20 a month for up to 10 people, $50 for up to 30, and $150 for up to 100, with yearly plans at $200, $500 and $1,500, which works out to roughly two months free. Larger organisations get custom pricing. Inside, a team can switch between OpenAI, Claude, Gemini and open-source models from one window, while admins get SSO, audit trails, PII detection and budget caps per user, department or project. There is also an open-source edition for teams that would rather run everything in their own cloud.
The math for a 30-person team: $50 a month for the platform. Then set a usage cap, say $10 per person per month. Even if every single person hit that cap, the worst case is $350 a month. The same 30 people on annual Standard seats would cost $600 a month. Your heaviest users might need a higher cap, but most of the team will never come close.
Great for: Growing teams with uneven usage, companies that want several AI models without several subscriptions, and anyone nervous about staff pasting company data into personal AI accounts.
Watch out for: You now manage two bills, the platform and the API usage. Someone needs to create the API accounts and set sensible limits. And as a newer, smaller vendor, it has fewer native integrations than Microsoft or Google, so check that it connects to the tools your team relies on.
All five models, side by side
| Model | Example | Starting price | Predictable bill? | Fair to light users? | Setup effort |
|---|---|---|---|---|---|
| 1. Flat seat | ChatGPT Business, Claude Team | $20–25 per user/month | Very | No | Minimal |
| 2. Two-speed seat | Standard + Premium seats | $20–125 per user/month | Mostly | Partly | Low |
| 3. Suite bundle | Gemini in Workspace, Microsoft 365 Copilot | $14–21 per user/month, plus suite | Very | No | Low if you already use the suite |
| 4. Pay-as-you-go | Direct API access | Per token | No | Yes | High |
| 5. Platform fee + your keys | Grengin | $20–150 per team/month, plus usage | Yes, with caps | Yes | Moderate |
No row wins everywhere. The first three trade fairness for simplicity. The fourth trades simplicity for fairness. The fifth tries to keep most of both, at the cost of a little setup.
The hidden costs no pricing page shows you
Whatever model you choose, these are the costs that tend to surprise teams in month three.
1. Idle seats. The single biggest leak in per-seat plans. People join, try it, drift away, and keep their licence for a year.
2. Base licences. Suite add-ons only work on top of the suite. If half your team does not already have the right Microsoft or Google plan, the real per-person cost can nearly double.
3. Premium creep. Two-speed plans start with five Premium seats and end with fifteen, because upgrading feels easier than saying no.
4. Usage limits and overages. Most flat seats have fair-use caps. Hit them, and you are either throttled or buying extra credits.
5. Admin time. Every model needs someone to add and remove users, review usage and answer “why can’t I access this?” Usage-based models need a bit more of this, at least in the first month.
6. Lock-in. A tool tied to one AI lab ties you to that lab’s roadmap. When a better model launches elsewhere, switching means a new contract, new training and moving everyone’s saved work.
7. Shadow AI. The most expensive cost of all, and the one that never shows up on an invoice. If the approved tool is too limited or too pricey to give to everyone, people quietly paste company data into personal AI accounts. That risk does not appear in any budget until something leaks.
The one question that decides it
Forget the headline prices for a moment and ask: how evenly does my team use AI?
| If your usage looks like… | The model that usually fits |
|---|---|
| Everyone uses it heavily, every day | Flat seat (Model 1) |
| Most people light, a few very heavy | Two-speed seat (Model 2) |
| Work lives entirely in Google or Microsoft | Suite bundle (Model 3) |
| You are building AI into your own product | Pay-as-you-go API (Model 4) |
| Uneven usage, growing headcount, want several models | Platform fee + your own keys (Model 5) |
The even-usage team should happily pay the seat tax. It buys simplicity. The uneven team, which describes most teams, should at least run the numbers on a usage-based setup before renewing.
Five mistakes teams make when buying AI
1. Buying seats for everyone on day one. Start with the people who asked for it. Expand when usage proves the demand.
2. Comparing headline prices instead of total cost. A $21 add-on that needs a $14 base plan is a $35 tool. A $50 platform fee with your own keys is a $50 tool plus usage. Compare totals.
3. Picking a model before picking a pricing model. Teams argue for weeks about which AI is smartest, then sign a contract that locks them into one. Model rankings change every few months. Your contract does not.
4. Ignoring the annual discount. Most vendors knock 15 to 20 percent off for paying yearly. Once a pilot proves the tool sticks, take it.
5. Never looking at the usage report. Every serious AI tool has one. Check it monthly. It is the only honest answer to “is this worth it?”
How to run a 30-day pricing pilot
- Pick 10 to 15 people from different roles, including a couple of heavy users and a couple of sceptics.
- Give them the tool you are considering, on monthly billing so you are not locked in.
- Track three numbers each week: who logged in, how often, and what it cost.
- Ask one question at the end: what would you miss if we turned this off tomorrow?
- Match the usage pattern to the table above, and choose the pricing model that fits the shape you actually saw.
Thirty days is short enough to stay cheap and long enough to get past the novelty bump, when everyone tries the new tool once.
The takeaway
AI pricing is still settling. In 2026 alone, the big labs cut some seat prices, added premium tiers costing five times more, and pushed large customers toward usage billing. Meanwhile, newer players are betting that teams want flat, predictable platform fees and the freedom to pick their own models.
The vendors are clearly still experimenting. So should you. Run the pilot, read the usage report, and let your team’s real behaviour, not the pricing page, pick your plan.
