Cognition is making a clearer case for Devin, but it is also making the real cost of agentic coding harder to ignore by pushing buyers toward quota, overage, and run-frequency decisions.
The easiest way to misunderstand Devin's new pricing is to focus on the cheapest number in the lineup.
The more useful way is to ask what new decisions the pricing model forces onto the buyer.
Cognition's April self-serve update matters because it changes Devin from a product you compare mostly by access price into a product you compare by control surface. Free, Pro, Max, Teams, and Enterprise are only the visible layer. Underneath that sits the real question: how often will Devin run, how much work will it do, and who decides when the expensive modes are worth it?
That is why this belongs in the same conversation as Cursor's metered review model and GitHub Copilot's usage-based billing shift. The coding-agent category keeps moving away from simple seat logic and toward workload-shaping logic.
Cognition says it is retiring the old Core and Team plans and replacing them with Free, Pro, Max, Teams, and Enterprise. Pro starts at $20 per month. Max sits at $200 per month. Teams becomes usage-based with an $80 monthly minimum, much lower than the older $500 team entry point.
On the surface, that makes Devin easier to try.
But the more important line in the announcement is the one about usage beyond included quota. For self-serve customers, that extra usage gets billed in dollars rather than the older ACU framing. Cognition is also starting to charge more explicitly for Ask Devin, Devin Review, and higher-cost DeepWiki modes, while promising better controls over when those products run.
That last part is the story.
If a tool only runs when a person explicitly asks for help, cost tends to feel manageable. Once the tool starts reviewing pull requests, generating deeper wiki artifacts, or operating on a schedule, the expensive part is not the headline plan. It is the frequency of autonomous work.
Cognition seems to understand that. The announcement talks about better control over when Devin Review runs, including manual only, run when a PR is first opened, or run on every commit.
Those are not tiny product settings. They are budget policy.
A team that runs review on every commit is making a very different financial bet from a team that runs review only when a PR first opens. That difference can matter more than whether the base plan starts at twenty dollars or two hundred.
Earlier Devin pricing discussions focused heavily on ACUs because they made the cost of active work feel explicit. That framing created its own anxiety: buyers could picture a compute meter spinning every time the agent kept working.
The new structure does not erase that concern. It just translates it into quota and dollar overage language that may feel friendlier while still asking the same operational question: do you understand the workload shape you are about to automate?
Butler has made the same point in other pricing coverage, including high-intensity coding-team pricing pressure and the review bottleneck behind coding-agent fatigue. The cheapest-looking tool can become the least predictable tool once autonomous work expands faster than review discipline.
Devin should not be compared only by entry price.
Buyers should compare four things instead.
If the included allowance covers normal use, the tool may feel straightforward. If not, the conversation quickly becomes about overages and behavior limits.
A vendor that lets you decide manual-only, first-open, or every-commit behavior is really giving you cost-governance tools. Those controls deserve as much attention as benchmark scores.
Lowering the team entry point can help experimentation, but it can also make it easier to expand before anyone knows what steady-state spend will look like.
Even a well-priced run can become expensive if humans still need to inspect too much generated work. That remains the quiet tax across the whole category.
The real question is not whether Devin is cheap or expensive.
It is whether your team is mature enough to govern a quota-based coding agent without drifting into surprise usage. Cognition is giving buyers more flexibility, which is good. It is also removing the excuse that nobody could tell when autonomous work was being triggered or how often it should happen.
That makes Devin more buyable and more demanding at the same time.
In this market, that is usually what product maturity looks like.
This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.