Plans, usage, and what an ACU actually is

Agent pricing is harder to reason about than seat pricing, because you are buying work rather than access. Here is what Cognition publishes — and, just as usefully, what it does not.

Self-serve plans

Verified against devin.ai/pricing on 28 July 2026:

Plan	List price	Members
Free	$0/month	1
Pro	$20/month	1
Max	$200/month	1
Teams	from $80/month	Unlimited

Pro and Max are individual plans and are documented as not shareable. Teams starts at an $80/month minimum, with each member taking either a full seat at $40/month or a flex seat.

How usage is metered

Each paid plan carries a usage allowance that refreshes on a daily and weekly cycle. Go past the included usage on a paid plan and you purchase extra usage at API pricing — that is, at the underlying model provider’s rates, not at a Devin-published per-token schedule.

One mechanic is worth knowing because it saves money quietly: when a session goes idle, Devin goes to sleep and stops consuming usage. The docs put the threshold at roughly 0.1 ACUs of inactivity. You are not billed for a session you walked away from and forgot to close.

Enterprise

Enterprise is billed in Agent Compute Units (ACUs) at a rate set in the order form, with pricing through sales rather than a public page.

What an ACU is — and what Cognition won’t say

The documented definition is about effort, not time. From the Desktop usage docs: an ACU “reflects the amount of agent effort required to complete a given task,” and “ACU consumption scales with the inference used and the model selected.” Elsewhere ACUs are described as representing “how much actual work the agent is doing.”

Enterprise ACUs are documented as distinct from self-serve quota and on-demand credits, tied to a mode that “adheres more strictly to task planning and end-to-end testing.”

Now the honest part. You will find blog posts and forum answers confidently converting ACUs into minutes — commonly around 15 minutes per ACU. That number is not in Cognition’s documentation. We could not source it, so we are not going to repeat it as fact. If your budgeting depends on a time conversion, get it from Cognition in writing rather than from a secondary source.

Three other things are also not publicly documented, and we flag them rather than guessing:

• the exact quota amounts for each plan (Max is documented as having a “significantly larger weekly usage quota” than Pro, without numbers)
• the per-credit price of on-demand credits beyond “consumed at API pricing”
• any published ACU-to-wall-clock conversion

What this means if you are evaluating Devin

Because overage lands at underlying API pricing, your effective cost tracks the model you select and how much inference a task needs — not a flat per-seat number. Two teams on identical plans can spend very differently.

That is an argument for the scoping discipline in Writing tasks Devin can actually finish: tight, verifiable tasks are not only more reliable, they are cheaper.

Prices and mechanics change. This guide records what the sources said on 28 July 2026 — check the pricing page before making a commitment.

References

1. Devin pricing — devin.ai
2. Self-serve billing — docs.devin.ai
3. Enterprise billing — docs.devin.ai
4. Usage — docs.devin.ai

From Devin Central — subscribe for Devin news, features & tutorials. Devin is a trademark of Cognition.