Per-seat vs usage pricing for AI coding agents: a team of 5

Most per-seat vs usage pricing comparisons for AI coding agents stop at the sticker price and skip what each model charges for. Seat prices follow headcount and credits follow how much the agent does. Sinatra sells a third model: one flat price for the whole workspace, which changes only when you need more work running at once. If you're the founder or CTO who reads the pricing page before the feature list, the useful exercise is one five-person team taken through all three, first in month one and then in month six, when the agent is doing three times the work.

We build Sinatra, which is priced the third way, so read the Sinatra parts as a vendor describing itself. Every competitor figure links to the vendor's own page. Where the token cost of a task would be a guess, we leave it out. Our startup cost post does that token math for a ten-person team, with the assumptions written down.

Product details checked October 2026 against each vendor's own pricing and billing pages, linked throughout.

The team

The team is five people. Three engineers hand tickets to an agent and review what comes back. A tech lead reviews most of the pull requests and rarely starts a task. A product manager writes the tickets.

In month one the three engineers are the only ones starting tasks. By month six the agent is doing three times as much: the engineers delegate more, the lead hands over the small fixes, and the PM has started assigning tickets straight to the agent instead of to a person.

The same two questions apply to every vendor below. What does it cost to let the lead and the PM in? And what happens to the bill when the work triples?

Per-seat pricing for coding agents

These team plans include a hosted coding agent in the seat. Here they are priced for five people:

Plan	Price per seat	Five seats	What the seat includes
Claude Team, Standard	$25 a month, or $20 billed annually	$125, or $100	Claude Code, with usage limits per person
ChatGPT Business, Standard	$25 a month, or $20 billed annually	$125, or $100	Codex, with included usage per seat
Cursor Teams, Premium	$30 a month	$150	Cloud Agents, with a usage pool per seat for Cursor's own models
Factory Teams	$60 a month plus $40 a seat	$260	Rate limits per seat, up to 10 seats

The seat is per person, and so is the usage inside it. Anthropic says Team usage limits "are per-member" and "can't be transferred or shared between team members." Cursor says "Seats don't share weekly pools with each other." OpenAI evaluates included usage "for the paid ChatGPT seat assigned to each member," and only after that can work draw on a workspace credit pool, if the workspace has flexible pricing turned on.

For our team, the lead and the PM each cost a full seat to start even one task, and most of that seat's usage sits idle all month. A few vendors now offer a way around it. Cursor Teams has a free seat with no usage allowance of its own, plus unpaid admins for staff "who don't need Cursor access," per its members page. Devin Teams has free flex seats that "draw entirely from the team's shared pool of on-demand credits," per its billing docs. Devin's full seats are priced two ways, though. The pricing page says "$80/month for team plan + $40/mo per full dev seat," while the billing docs describe an $80 monthly minimum and $40 a full seat. Ask Cognition before you budget. Linear and ChatGPT Business go the other way: Linear guests "are billed as regular members," and ChatGPT charges for every seat "whether or not they've been assigned."

When usage triples, a seat plan's bill doesn't move until someone hits their limit. Then that one person moves to a bigger seat: Claude's Premium seat is $125 a month, ChatGPT's is $125 a month and drops the 5-hour limit, and Cursor's Super and Ultra seats are $100 and $200.

Plenty of teams also pay for these seats already. If your five people are on Claude Team or ChatGPT Business for chat anyway, the coding agent inside the seat costs nothing new. Anthropic says cloud sessions count toward your plan's usage limits and "there's no separate charge for the cloud VM."

Usage-based pricing: credits and balances

Credit and balance pricing charges for what the agent spends, and it's usually sold alongside a per-person plan rather than instead of one.

GitHub Copilot is the clearest case. Copilot Business is $19 per granted seat, and each seat brings 1,900 AI credits a month at $0.01 a credit. The cloud agent draws on those credits per token. Credits also pool: GitHub's billing docs say "A user's included AI credits are pooled at the billing entity level," so a light user's credits cover a heavy one. Five seats are $95 a month with 9,500 credits in the pool. Unused credits don't carry over to the next month. Our Copilot pricing post goes through how fast agent work uses them.

Linear's coding sessions are almost pure usage. Linear itself is $10 a user a month on Basic and $16 on Business, billed yearly, so $50 or $80 for five. Sessions come out of a prepaid workspace balance at, per its AI credits page, "model tokens at provider-published rates, with no markup, and sandbox runtime at $0.25 per 20-minute block." Those are published rates, so once you know roughly how many tokens and minutes a session takes, you can work out its cost. The Linear coding sessions explainer has the rest.

Cursor and Devin mix seats and usage. A Devin full seat comes with an allowance, and past it, usage comes out of credits the whole team shares. On Cursor Teams the seat's pool covers only Cursor's own models. Third-party models such as Claude are billed at API list price plus a Cursor Token Rate, "paid from the team's prepaid balance from the first request," per its Teams pricing page, and the token rate of $0.25 per million tokens applies on your own key too. Our Cursor post has more on how that works.

Letting the lead and the PM in costs little here. On Copilot each adds a seat's worth of credits to the pool, and on Linear each costs a user license and nothing more until they start a session. As the work grows, so does the bill: triple the work and the usage line roughly triples with it, with no ceiling unless someone sets a budget.

Flat-rate pricing per workspace

The third shape charges one price for everyone and limits something other than people. Sinatra prices this way. So does Augment Code: "One flat price for your whole team, up to 50 seats with $20 of usage included every month," plus what its pricing page calls "a flat 40% fee on LLM usage."

Sinatra's plans are priced by how many tasks run at once. Free is 1 task at a time, plus 1 queued, with no daily limit and no card. Pro is $20 a month for the whole workspace: 3 tasks at a time, plus 10 queued, any number of people. Team is $99 for 10 tasks at a time, plus 25 queued, and Scale is $249 for 25 tasks at a time, plus 100 queued. Work past the limit is queued, and nothing caps tasks per day or month. We dropped per-member pricing on October 12, so inviting a reviewer no longer costs anything.

The plan doesn't include any model usage. Every member who starts a task needs model access: their own Claude, ChatGPT or SuperGrok subscription, or an Anthropic or OpenRouter API key an owner or admin adds for the workspace. A personal subscription covers only the person who connected it. Sinatra never bills for model usage. That's between you and your provider. So on a flat plan the model cost is still per person if you use subscriptions, or per token if you use a shared key. The BYOK post weighs the two.

Month one and month six, side by side

	Letting in the lead and the PM	Work triples by month six	What stops a runaway month
Per seat (Claude Team, ChatGPT Business, Cursor Teams, Factory)	Two more seats, unless a free or flex seat type covers them	Nothing, until someone hits their limit and needs a bigger seat	Each person's usage limit
Usage (Copilot, Linear sessions, Cursor and Devin past the allowance)	A license each; Copilot adds their credits to the pool	The usage line roughly triples	A budget, if someone sets one
Flat workspace (Sinatra, Augment)	Nothing on the plan; on Sinatra, each needs model access	Nothing on the plan; more work is queued or needs a bigger plan	The plan's limit on work at once

Sinatra in month one: three engineers on subscriptions they already pay for, on Pro, is $20 a month. The lead and the PM join for nothing on the plan. The PM, who has no coding subscription, starts tasks on the workspace API key, and those tokens land on your provider account at its rates.

Sinatra in month six: the plan stays at $20 as long as three tasks at a time is enough. When it isn't, tickets are queued and start as running ones finish. If tickets are queued most of the day, Team is $99 for 10 tasks at a time, plus 25 queued. The engineers' subscriptions carry three times the work, which uses more of each subscription's limits and leaves the price where it was. If they hit those limits, the fix is a bigger subscription or the shared key. For this team, Sinatra costs the subscriptions you already pay for, plus $20 a month for the whole workspace.

Where seats and credits win

If your team already pays Claude Team or ChatGPT Business seats for everyone, the coding agent in those seats is the cheapest option in this post, because you've already paid for it. Sinatra would run on subscriptions like those rather than replace them.

Pooled credits suit uneven teams. On Copilot, one heavy user can draw on the credits of four light ones, which per-person limits on Claude Team and Cursor don't allow.

Linear bills tokens at the providers' published rates and sandbox time at a published $0.25 per 20 minutes, so a team that tracks what its sessions use can forecast spend from those rates.

Free reviewer seats remove most of the seat penalty. On Devin Teams the lead and the PM can sit on free flex seats, and Cursor Teams has a free seat. Check whether your vendor has one before you rule out seat pricing.

Augment's flat plan includes $20 of usage, and Sinatra's includes none.

Common questions

Is per-seat or usage pricing cheaper for a small team?

It depends on who uses the agent. If everyone already pays for a seat that includes one, the seat is cheapest. If two or three people do most of the delegating and the rest only review, pooled credits or a flat workspace plan can cost less, because the reviewers don't each need a seat.

Do reviewers need a seat?

On Claude Team, ChatGPT Business and Linear, everyone on the plan pays for a seat. Cursor Teams has a free seat type and unpaid admins, and Devin Teams has free flex seats that draw on shared credits. Reviewing the pull request on GitHub itself only needs access to the repository. On Sinatra, nobody pays to join the workspace.

What does "flat rate" leave out?

Usually the model. Sinatra's plan covers the sandboxes, the workflow and the dashboard, and model use goes to your own subscription or API key. Augment's flat plan includes $20 of usage, and its 40% fee applies to all LLM usage. Ask any flat-rate vendor what happens to the model bill before you compare prices.

What happens on Sinatra when more tasks come in than the plan runs at once?

They're queued and start on their own as running tasks finish, and a full queue is reported on the ticket. The billing docs say what counts and what doesn't.

The pricing page has every plan, and startups can get Team at half price for a year: $49.50 a month for 10 tasks at a time, plus 25 queued. For every vendor's meter side by side, see AI coding agent pricing compared. To try the flat plan on your own subscriptions, start free.