Devin vs Claude Code in 2026: the $20 tier is an editor now (field log)

Quick answer

Devin vs Claude Code in 2026 is no longer the "autonomous cloud agent versus terminal agent" fight most of page one still describes. That framing was right in 2025. Today Devin's $20 Pro tier is mostly an editor: the pricing page sells "Unlimited inline edits" and "Unlimited Tab completions" first, and lists "Access to cloud agents (Devin Cloud)" as one bullet among several (Devin pricing, 2026). Claude Code went the other direction. It stayed one agent, added surfaces, and still "asks for permission before making changes to your files or running commands" (Claude Code, 2026). The useful 2026 question is not which one is more autonomous. It is whether you want a multi-model editor that can dispatch cloud agents, or a Claude-only agent you supervise.

I did not run Devin for a week to write this. I sat down with both pricing pages and Devin's billing docs open side by side, because two of the claims that everything on this SERP repeats turned out to be out of date. That was the more useful exercise.

What Devin actually is in 2026

Devin launched as the "AI software engineer" you assigned a ticket to and walked away from. Cognition then bought Windsurf, and the product lines merged. You can see the merge in the plan sheet.

The Free tier gets "Light quota to code with agents," limited model availability, unlimited inline edits and unlimited Tab completions. No Devin Cloud. Pro at $20 a month adds "Increased quotas, including access to OpenAI, Claude, and Gemini frontier models," free use of SWE 1.7, cloud agents, and the option to "purchase extra usage at API pricing." Max at $200 a month is the same thing with much bigger quotas. Teams starts at $80 a month plus $40 per full developer seat, with unlimited flex seats on top (Devin pricing, 2026).

The tell is in the docs, not the marketing. Devin's own billing page describes the Pro quota as one that "covers Devin sessions, Devin CLI, and Devin Desktop," and the link behind Devin Desktop points at windsurf.com (Devin docs, 2026).

So in 2026 Devin is a multi-model IDE with Tab completion that also happens to dispatch cloud agents. That is a different product from the one every comparison page on this query is comparing.

What Claude Code actually is in 2026

Claude Code stayed narrow on models and got wide on surfaces. Anthropic lists it in the terminal, VS Code, JetBrains, the browser, iOS, Android, GitHub Actions and Slack. It can execute "across 10s to 100s of parallel subagents, and checking its work before anything reaches you." The default posture has not moved: it asks before it writes files or runs commands (Claude Code, 2026).

Pricing is a flat subscription rather than metered work. Free is $0 with no Claude Code. Pro is $17 a month annual or $20 monthly. Max starts at $100 a month, in 5x and 20x usage tiers. Team is $20 per seat annual or $25 monthly, with $100 premium seats at 5x usage (Claude pricing, 2026).

One axis is genuinely lopsided and worth saying plainly. If you want to run a non-Claude model, Claude Code is not the tool. Devin Pro hands you OpenAI, Claude and Gemini frontier models on the same $20. That is a real advantage and I am not going to bury it.

The ACU thing everybody is still quoting

Here is the correction that made me write this up.

Search around and you will find Devin discussed in ACUs, Agent Compute Units, usually with a per-ACU dollar figure attached and a $500 minimum. devin acu still pulls about 70 US searches a month on its own (DataForSEO, August 2026). That vocabulary is now Enterprise-only.

Devin's billing docs are explicit about the split. "Enterprise customers consume Agent Compute Units (ACUs) against the volume in their order form. Self-serve customers consume their plan's included quota first, then draw from prepaid on-demand credits" (Devin docs, 2026).

And the migration note is blunter than I expected:

> On-demand credits are the same dollar value as the ACUs you're used to. Legacy Core plan users have been migrated to the Free plan and can continue using any remaining on-demand credits.

Read that second sentence again. If you were on the old Core plan, you are on Free now. Any comparison page still quoting a self-serve ACU rate is describing a plan that does not exist. On-demand credits roll over month to month and never expire, which the old model did not do either (Devin docs, 2026).

What Devin does not bill you for

This is the part I think actually decides routing, and I have not seen it in any of the comparisons ranking for this query.

Devin's usage docs list what accrues: the "number and complexity of actions Devin takes (planning, context gathering, task execution, browser actions, code execution, and so on)," plus VM time and bandwidth, which the docs call "typically a small fraction of total usage." Then comes the list of things that do not accrue. Devin does not consume usage while waiting for your response, while waiting for a test suite to run, or while setting up and cloning repositories. Idle sessions sleep automatically "after roughly 0.1 ACUs (or the equivalent quota / on-demand credit) of inactivity," and sleeping costs nothing (Devin docs, 2026).

There is also an oddly specific line worth knowing if you work on a Windows target: Windows sessions consume approximately 9% more usage than equivalent Linux sessions.

Put those together and the intuition inverts. The mental model most people carry is "the autonomous one runs unattended, so it burns money unattended." It does not. Devin bills machine actions, and waiting is not an action. What it bills hard is churn: long conversations, big context, many files touched, lots of back and forth. The docs say so directly in their own tips section, which advises delegating "clearly scoped tasks with a well-defined end goal" and splitting big projects across sessions.

Claude Code bills the opposite thing, in effect. The subscription is flat, so machine time is free to you, but the scarce resource is your attention. You are sitting there watching it.

That is the decision rule, and it has nothing to do with which one is smarter.

The demand curve says people are only now asking

Small aside, because it changed how I read the SERP. US search volume for devin vs claude code ran about 90 a month in December 2025 and about 590 by July 2026 (DataForSEO, August 2026). devin ai pricing moved the same way, roughly 70 to 590 over the same stretch.

So most of the ranking pages were written before the question got popular, and a good number of them before the Windsurf merge landed in the plan sheet. That is why the frame is stale in so many places at once. It is not that anyone is being careless. The product moved underneath the articles.

How I would route work between them

• Long, verifiable, well-scoped tasks where I will not be watching: Devin cloud agents. Waiting is free, and a clean spec is exactly what its billing rewards.
• Anything where a wrong turn is expensive to unwind: Claude Code, for the ask-first default.
• Work that needs a non-Claude model: Devin, because Claude Code is Claude only.
• Exploratory work with lots of back and forth: Claude Code. That shape is precisely what Devin's own docs tell you burns quota fastest.
• Tab completion in an editor all day on $20: Devin, which is a sentence I could not have written a year ago.

The honest caveat is that I am reading meters, not stopwatches. Neither vendor publishes a task-level cost table you could plan a month around, and I am not going to invent one. If you are choosing on budget, run one real ticket through each and read your own invoice.

If you want the wider set, I keep a running list of the AI coding agents I actually reach for, and the Cursor comparison covers the editor-shaped end of this more closely than this one does.

p.s. I went in expecting to write about autonomy and came out writing about invoices. That keeps happening.