Pricing guide
Devin Pricing (2026): Devin AI Cost Per Month, Plans, Team Seats and ACU Costs
Every current Devin plan, what a team actually pays per seat, and why the ACU figures still quoted across the web no longer match what Cognition sells. All prices read from Devin's own pricing page in August 2026.
No card to start · Review-first: the agent never merges
Pick a task
Plan
- planning
Files changed
Test run
Pull request
You review and merge. Agentcode never merges on its own.
The short answer
Devin pricing in September 2026 is $0 for Free, $20 a month for Pro and $200 a month for Max, with Teams at $40 per full developer seat against an $80 a month minimum and Enterprise priced on request. The important change is that self-serve plans no longer bill in ACUs: Cognition retired the $500 team plan and the $2.25 Agent Compute Unit, and each paid plan now carries a usage quota that refreshes daily and weekly, with extra usage bought at API pricing. The $80 is a floor rather than a base fee, so once you hold two or more full seats the bill is simply seats times $40 and a ten-developer team pays $400 a month before any overage, and the quota size per tier is not published in units you can forecast. The flat-price way to run an agent on your repo, with no quota at all, is Agentcode at $29 a month billed yearly.
Devin plans and prices
Five plans: three for individuals, two for organizations. The column that decides your real cost is the quota, because that is the allowance agent runs draw from before you start buying extra usage.
Swipe to see more
| Plan | Price | Included usage | Best for |
|---|---|---|---|
| Free | $0 free | Light quota, limited models | Trying an agent on a real repo. |
| Pro | $20 per month | Increased daily and weekly quota | One developer delegating real tasks. |
| Max | $200 per month | Significantly higher quota | A power user running the agent all day. |
| Teams | $80 per month + $40 per full dev seat | Per-seat quota, unlimited sessions | A team that needs admin and shared billing. |
| Enterprise | Custom contact sales | Negotiated | A larger or regulated organization. |
Prices as published by Cognition on devin.ai and checked in August 2026. Cognition has repriced Devin more than once, so confirm on their site before you buy.
Does Devin still charge per ACU?
No, and this is the single most misreported fact about Devin pricing. If you are budgeting from a number you read in a comparison article, check the date first.
The retired model
ACUs, at $2.25 each
Devin originally metered work in Agent Compute Units. One ACU stood for roughly 15 minutes of active autonomous work, extra ACUs sold at about $2.25, and the team plan cost $500 a month with 250 ACUs included. That is the model nearly every "Devin pricing" article on the web still describes.
What Cognition publishes today
A quota, then API pricing
The current pricing page lists no ACUs at all. Each paid plan carries a usage allowance that refreshes automatically on a daily and weekly basis. If you exceed it, you buy extra usage, which is consumed at API pricing. Cost per message varies with the model, the size of the task and how much reasoning it needs.
The practical consequence is that Devin's billing now looks much more like Claude Code's than like a metered cloud service: a subscription with rolling quota windows rather than a unit you can count. That is friendlier for most individual users, who will never hit the ceiling. It is harder for a finance team, because Cognition does not publish the size of each tier's quota in any unit you can forecast against a sprint. You find out where the ceiling is by hitting it.
What each Devin plan gets you
Free
$0 · freeA light usage quota with limited model availability, plus unlimited inline edits and unlimited Tab completions in Devin Desktop. Enough to see how the agent scopes a task and what its output looks like on your own code. Not enough to do professional work through.
Pro
$20 · per monthThe plan most individual buyers land on. It raises the quota and opens full model availability, including frontier models from OpenAI, Anthropic and Google, free use of Cognition's own SWE 1.7 and leading open source models, and access to cloud agents through Devin Cloud. Extra usage past the quota can be purchased at API pricing.
Max
$200 · per monthEverything in Pro with a significantly larger usage quota. This is the tier for someone whose working day is mostly agent runs rather than hand-written code, and who would otherwise buy a lot of extra usage on top of Pro. It is still a single-member plan.
Teams
$80 · per month + $40 per full dev seatAn $80 monthly base plus $40 for each full developer seat. Team members themselves are unlimited and free: flex seats cost nothing, and you only pay $40 for someone who needs a full quota of their own plus Devin Desktop. Adds sharing and collaboration, centralized billing, an admin dashboard with analytics, the Devin API, unlimited concurrent sessions and priority support.
Enterprise
Custom · contact salesEverything in Teams plus deployment in your own virtual private cloud, SAML/OIDC SSO, centralized admin controls across multiple Devin organizations, teamspace isolation, custom git provider support, custom terms and a dedicated account and engineering team. Priced on an order form rather than published.
How much does Devin cost for a team?
Devin Teams is a base fee plus a per-seat fee, which is the part that catches people out: the headline $80 is not the price of a team, it is the price before anyone has a seat. Here is what the formula produces at real headcounts.
Swipe to see more
| Team size | Calculation | Monthly total |
|---|---|---|
| 1 full seat | $40 seat + $40 of prepaid credits | $80 / mo |
| 2 full seats | 2 x $40, minimum exactly met | $80 / mo |
| 5 full seats | 5 x $40 | $200 / mo |
| 10 full seats | 10 x $40 | $400 / mo |
| 25 full seats | 25 x $40 | $1,000 / mo |
| 50 full seats | 50 x $40 | $2,000 / mo |
One nuance in Cognition's favor that most write-ups miss: members on a Teams plan are unlimited and free. Flex seats cost nothing, and the $40 applies only to a full developer seat, which is the one that carries its own quota and Devin Desktop access. So a twelve-person team where four people actually delegate work to the agent pays $80 plus four seats, not twelve. Budget by how many people will genuinely run agent tasks, not by headcount.
Totals above exclude extra usage. They are the floor, not the forecast. If your team pushes past the included quota during a heavy sprint, the additional usage bills at API pricing on top of every figure in the table. For how this shape compares across the whole category, see the flat versus metered AI coding tool pricing breakdown.
What happens when the quota runs out
This is the half of Devin's pricing that decides a heavy month, and it is not on the pricing page. Extra usage is sold as on-demand credits, and Cognition's own documentation answers the question everyone arrives with: how do these relate to the ACU figures in every article written before the repricing? Verified against Cognition's documentation in August 2026.
Swipe to see more
| What | Answer | Why it matters to a buyer |
|---|---|---|
| What replaced ACUs | On-demand credits | Cognition publishes the conversion in its own words: "On-demand credits are the same dollar value as the ACUs you're used to." So the ACU figures still quoted in older articles are not meaningless, they are just the previous name for the same dollar of usage. That is the sentence to keep if you are reconciling an old quote against a current one. |
| Do credits expire? | No | The docs are explicit that purchased credits "never expire" and "roll over month-to-month". That is unusual. Most agent vendors reset the balance every month, so a quiet month is money lost. Here it is not, which materially changes how you should buy: top up in larger, less frequent amounts rather than trying to match each month. |
| How you top up | Settings, then Plans | Credits "can be topped up at any time from Settings > Plans, and optionally refilled automatically via auto-reload". Auto-reload is the setting to think about before you switch it on, because it is the one that turns a usage spike into an invoice without anyone approving it. |
| What happens at zero | Two things stop | While you have credits, "on-demand credits keep you working without interruption". At zero, the docs single out two features that degrade rather than queue: "Automations stop running and Devin Review switches to its smart diff viewer." If either of those sits in your pipeline, a balance of zero is an outage in that pipeline, not just a pause. |
| Flex seat, in detail | Free, unlimited | A flex seat carries "no fixed monthly charge per seat" and you can have unlimited ones, but read the two conditions: flex seats "draw entirely from the team's shared pool of on-demand credits" and they "do not include Devin Desktop access". So free members are free in seat cost only. Their work still spends the team pool. |
| Full seat, in detail | $40 / mo | A full seat buys "a daily and weekly usage quota equivalent to the Pro plan" plus Devin Desktop. The useful way to read that: a full seat is a $20 Pro plan plus Desktop and admin, sold at $40 inside a team. Buy full seats for the people whose usage is constant and leave everyone else on flex. |
Rolling credits that never expire are genuinely unusual and they change the buying advice. On a vendor whose balance resets monthly, you buy small and often. Here you can buy once and let the balance sit. The trade is that nothing forces you to notice consumption, which is the same structural problem GitHub Copilot code review has with its own credit meter: no rate limit ever tells you to slow down.
Which Devin plan do you actually need?
The tiers are separated almost entirely by quota size rather than features, so the decision is about how much agent work you do, not which capabilities you unlock.
Free is for evaluation, not for work
The free tier exists so you can point the agent at a real repository and judge the output on code you know. That is a fair use of it. What it will not do is carry a week of actual tasks: the quota is light and model availability is limited, so you get a sense of quality and then you hit the wall. Treat it as a trial with no clock on it rather than a plan.
Pro at $20 covers most individual buyers
Pro opens full model availability and Devin Cloud, and Cognition's own guidance is that the included quota covers all agent usage for the majority of users. If you delegate a few well-scoped tasks a day and write the rest of your code yourself, $20 is very likely the whole bill. The people who outgrow it are the ones who stop writing code by hand.
Max at $200 is a bet against overage, not a feature upgrade
Max adds no capability over Pro. It is ten times the price for a significantly larger quota, which makes it worth buying only if you would otherwise purchase more than $180 of extra usage a month on Pro. The honest sequencing is to run Pro first, watch where you land against the quota, and move up if you are consistently buying usage rather than guessing in advance.
Teams is about administration as much as quota
What you buy at $80 plus $40 a seat is centralized billing, an admin dashboard with usage analytics, sharing and collaboration, the Devin API, unlimited concurrent sessions and priority support. If you have three developers expensing Pro individually, moving to Teams costs more but gives finance one invoice and gives you visibility into who is burning the quota. That visibility is worth real money once more than a couple of people are involved.
Devin vs a flat-priced PR agent
Devin and Agentcode are in the same category, which makes this a fairer comparison than most on this site. Both take a task and hand back a pull request. They differ on breadth and on how the bill behaves.
Swipe to see more
| Devin | Agentcode | |
|---|---|---|
| Billing model | Subscription with a usage quota that refreshes daily and weekly; extra usage bought at API pricing | Flat subscription, no usage quota and no overage |
| Entry price | $0 free, $20 / mo Pro | From $29 / mo billed yearly |
| Team pricing | $80 / mo base plus $40 per full developer seat | Flat per-plan price, unlimited repos on Team |
| What a busy month costs | More: heavy agent work runs past the quota and bills as extra usage | The same as a quiet month |
| Quota transparency | The size of each tier's quota is not published in units you can forecast | Nothing to forecast; there is no meter |
| Opens pull requests | Yes, and can be delegated from Slack, Teams, Linear, Jira, a CLI or the API | Yes, reviewable PRs on GitHub and GitLab |
| Git hosts | GitHub, GitLab and Bitbucket; custom providers on Enterprise | GitHub and GitLab |
| Surface area | Wider: desktop app, Tab completions, cloud agents, DeepWiki, an API for running agents programmatically | Narrow on purpose: connect a repo, review pull requests |
| Model choice | Yes, frontier models from several providers plus its own SWE 1.7 | Managed model selection, nothing to configure |
| Trains on your code | Enterprise controls and VPC deployment available | Never, on any plan |
Devin is the broader product and it is not close: a desktop app, Tab completions, delegation from Slack, Linear and Jira, Bitbucket support, DeepWiki and a real API for running agents programmatically. If you want an agent platform, buy the platform. Where a flat plan wins is predictability. Agentcode does one job, task to reviewed pull request on GitHub and GitLab, for one price that does not move when a sprint gets busy. For the capability-by-capability view see the Devin alternative page and the Agentcode vs Devin comparison.
Devin pricing: your questions answered
How much does Devin AI cost?
Devin costs $0 on the Free plan, $20 a month for Pro, and $200 a month for Max, as of September 2026. Teams is $40 per full developer seat with an $80 a month minimum, so a ten-developer team pays $400 a month. Enterprise is custom priced. Every paid plan includes a usage quota, and work past it is bought as on-demand credits.
Does Devin still charge per ACU?
Not on the self-serve plans. Paid plans now include a usage quota that refreshes daily and weekly, and extra usage is bought as on-demand credits. Cognition publishes the conversion itself: "On-demand credits are the same dollar value as the ACUs you're used to." So an old ACU figure still translates, it is just no longer the unit you are billed in.
What is an ACU in Devin?
An Agent Compute Unit was Devin's original metering unit, roughly 15 minutes of active autonomous work, sold at about $2.25 each alongside a $500 monthly team plan. That model is no longer what Cognition publishes. If you are comparing quotes or reading older coverage, check the date, because ACU-based figures no longer match the current plans.
Is Devin AI free?
Devin has a genuinely free tier. It gives you a light usage quota with limited model availability, plus unlimited inline edits and unlimited Tab completions in Devin Desktop. It is designed for evaluation rather than daily work: the quota is small enough that a serious task or two will exhaust it, which is what the $20 Pro plan is for.
How much does Devin cost for a team?
Devin Teams is $40 per full developer seat with an $80 a month minimum, not $80 on top of the seats. Five full seats cost $200 a month, ten cost $400 and twenty-five cost $1,000. Below two full seats the rest of the $80 is charged as prepaid credits the team can spend, and flex seats are free, so extra members cost nothing until they use credits.
What is the difference between Devin Pro and Max?
Both are single-member plans with the same feature set. The only meaningful difference is quota size: Pro at $20 a month includes an increased daily and weekly allowance, while Max at $200 a month includes a significantly higher one. Max pays for itself only if you would otherwise buy a lot of extra usage on top of Pro.
Is Devin AI worth it?
Devin is worth it when you want a genuinely broad autonomous agent: delegation from Slack, Linear or an API, a desktop app, Tab completions and cloud agents in one product. It is poor value if all you need is a steady stream of reviewed pull requests, because you are paying for surface area you will not use and carrying a quota you cannot forecast.
Does a new Devin Teams plan include SSO?
No. Cognition moved SSO out of the Teams plan and made it an Enterprise-only feature, so a new self-serve customer cannot buy it at any headcount. Teams who purchased SSO as an add-on before the change keep access under their grandfathered plan. If a security review requires SAML or OIDC, price the Enterprise order form rather than counting $40 seats.
Do Devin on-demand credits expire?
No. On-demand credits roll over month to month and purchased credits never expire, which is unusual in this category and is the strongest argument for prepaying. You can top up at any time or set an auto-reload threshold, and admins can set default session spending limits. On Teams the balance is shared across all members with no per-person allocation.
What do existing Windsurf subscribers pay?
Less than a new buyer, indefinitely. Cognition grandfathers an existing Windsurf Pro subscriber at $15 a month and an existing Teams subscriber at $30 per developer seat, against $20 and $40 for someone signing up today. There is no published end date for those rates. The editor itself is now Devin Desktop, and the detail sits on our <a href="/devin-desktop-pricing" class="link-merge">Devin Desktop pricing</a> page.
Is there a flat-price alternative to Devin?
Yes. Agentcode charges a flat fee from $29 a month billed yearly with no usage quota and no overage: you describe a task, it plans the work, edits the code, runs your existing test suite and opens a reviewable pull request on GitHub or GitLab, for the same price whether you run it once or fifty times. It is narrower than Devin by design, which is what makes the bill predictable.
Last updated: August 2026
Comparing Devin against the field?
These pages price the rest of the category the same way: Claude Code pricing, Cursor pricing, GitHub Copilot pricing, Replit pricing, and the Devin alternative breakdown.
Or step back to AI coding tool pricing across every tool, the best AI for coding in 2026, or what an AI coding agent does before you weigh the bill.
A flat bill, on your repo
Connect a GitHub or GitLab repository, describe a task, and read the pull request it opens. Same price whether you run it once or a hundred times, with no quota to watch.