Devin pricing: $20 to $200, and the $2.25 ACU is gone

How much does Devin cost on each plan?

Devin has four self-serve plans and an Enterprise contract, and each paid plan buys you a usage quota rather than a fixed number of tasks. We read every row from Devin's pricing page and its self-serve billing docs on 8 November 2026. US prices, before tax:

Devin doesn't list a yearly price, so every plan is monthly. Cognition doesn't publish the size of any quota either. Its pricing FAQ only says the allowance "refreshes automatically on a daily and weekly basis," and that "Most users will never hit their limits." Pro and Max are single-user plans. If two people need Devin on one bill, that's the Teams plan. The same plans also pay for Devin Desktop, which Cognition says "is the new name for Windsurf." This page prices the Devin cloud agent. The editor side, from Tab completions to its per-model rates, is on Windsurf pricing.

What happened to Devin ACU pricing?

Cognition stopped billing self-serve customers in ACUs on 14 April 2026. Its announcement retired the Core and Team plans and said "Usage beyond included quota will be priced in dollars based on the underlying model cost of each run." It also said "Enterprise billing will continue to use ACUs." Many pricing guides still quote the old rates. Here's what changed, side by side:

An ACU, or Agent Compute Unit, was Devin's own unit of work. The old pricing page called it "a normalized measure of the computing resources Devin uses to complete a task, such as virtual machine time, model inference, and networking bandwidth." On Core, the $20 minimum bought about 8.9 ACUs ($20 divided by $2.25). Core users moved to Free, and Cognition said Free doesn't keep the old no-minimum pay-as-you-go setup. Team users moved to Teams.

Enterprise ACUs aren't the old ACUs either. Devin's Enterprise billing page says they're "distinct from self-serve quota and on-demand credits" and "priced per the customer's Enterprise order form." So there's no public ACU price you can quote today.

What does a Devin team pay now compared with before?

A Devin team of two to twelve developers pays less than it did before April, and a team of thirteen or more pays more. The old Team plan was a flat $500 for any number of users sharing 250 ACUs. Teams now costs $40 per full seat, so the bill grows with the headcount. We priced each team size with every member on a full seat:

The allowances aren't the same thing, so this isn't a like-for-like price. The old $500 came with 250 ACUs for the whole team. A full seat now comes with its own Pro-sized quota, which Cognition doesn't measure in public. A big team that only uses Devin now and then can still pay less. It can put most people on free flex seats, which draw from shared on-demand credits instead of a seat fee.

How much does a Devin task cost?

A Devin task costs whatever its model and machine time cost, and Cognition doesn't publish a typical figure. Its usage docs say usage grows with the "Number and complexity of actions Devin takes," plus virtual machine time, which is "typically a small fraction of total usage." You can see what one session cost in Session Insights once it's done.

Some of the waiting is free. Apart from the few units it takes to keep the machine up, Devin uses nothing while it waits for your reply, waits for a test suite or clones a repo. It goes to sleep after 5 minutes idle, and a sleeping session costs nothing. Windows sessions use "approximately 9% more" than Linux ones, and macOS costs the same as Linux for now, under what Cognition calls "promotional launch pricing."

Past your quota, you pay the model's API price out of on-demand credits. Cognition doesn't print a rate card for those credits. As a guide, Anthropic lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens on its pricing page, read 8 November 2026. At those rates, $10 of usage is 5 million input tokens or 1 million output tokens. On-demand credits roll over from month to month, and the docs say "Purchased credits never expire."

What does a month of Devin cost?

One developer pays $20 a month on Pro until the quota runs out, and a team of five regular users pays $200. Here's how the common setups add up, using the prices on the plan page:

Max costs $180 more than Pro. So if you're buying more than $180 of on-demand credits most months, Max is probably the cheaper way to get the same work done. Cognition doesn't say how much bigger the Max quota is, so check the usage page after a heavy week before you switch. The ChatGPT row only covers GPT models. Devin's ChatGPT billing page says non-GPT models, fast GPT variants and "The machine time of Devin Cloud sessions" still use your Devin quota.

Is Devin free?

Yes. Devin Free costs $0 and includes "Limited Devin usage," Devin Review and DeepWiki, according to the self-serve docs. The pricing page calls the agent allowance a "light quota" on a limited set of models. You can't link a ChatGPT plan on Free.

Some of Cognition's tools stay free on any plan. Cognition said in April that Devin Review stays free on open source repos, and that deepwiki.com and its standard wiki generation stay free too. What isn't free since April: Deep Mode in Ask Devin, Devin Review on other repos after a 2-week free trial, and the pricier DeepWiki modes.

What does Devin cost for a team?

Devin Teams costs at least $80 a month for up to 200 people, and each regular user takes a $40 full seat. Below two full seats, Cognition charges the rest of the $80 as prepaid credits that the whole team shares. From two full seats up, you've met the minimum and pay $40 per seat. Flex seats are free and draw only from those shared credits. They don't get Devin Desktop.

Automations and Devin Review on Teams always spend the shared credits, not a full seat's quota. If the credits run out, the docs say "Automations stop running and Devin Review switches to its diff viewer." Past 200 people, or for SSO and a private cloud, you need Enterprise, which Cognition prices by contract in ACUs. Enterprise admins can cap that spend with budgets for the whole account, for pools and for each user.

What does Devin cost next to the rest of your AI stack?

Devin Pro plus ChatGPT Plus plus Claude Pro costs $60 a month, and the ChatGPT part can now pay for some of your Devin work. ChatGPT Plus is $20 on OpenAI's plan page. Claude Pro is $20, or $17 a month if you pay for a year, on Anthropic's pricing page. We read both on 8 November 2026. Claude Pro comes with Claude Code and ChatGPT Plus comes with Codex, so this stack holds three coding agents.

Whizi, which publishes this page, has no coding agent, no IDE and no API, so keep Devin for the work in your repo. Whizi Pro costs $29.99 a month for 2,500 credits and puts Claude, GPT and Gemini chat in one app, plus 100 pictures a month, with credits per message varying by model. Video and music sit on the Powerhouse plan. It can stand in for the chat side of a ChatGPT and Claude pair, but not for Codex or Claude Code, and it can't pay for Devin usage the way a linked ChatGPT plan can. For the other agents, see Codex pricing and Kiro pricing.