Devin Desktop Pricing in 2026: What the Cognition–Windsurf Merger Changed

In short: Devin Desktop is the renamed Windsurf IDE, now owned by Cognition, maker of the Devin coding agent. After the 2025 merger it dropped Windsurf's credit system for daily and weekly quotas: a Free tier, Pro at $20 a month, and a power-user Max tier at $200. Enterprise autonomous work still bills in Agent Compute Units.

If you opened your editor one morning in June and found Windsurf calling itself Devin Desktop, you were not imagining it. The name changed, the price changed, and the credit meter you used to watch vanished. None of it was random.

Why does Devin Desktop pricing feel so confusing right now?

The confusion has a simple root: three changes landed inside twelve months, and they arrived out of order for most users. First the price moved. Then the credit system disappeared. Only months later did the app itself get a new name. By the time the Devin Desktop rebrand timeline reached your update channel on June 2, 2026, the pricing had already been reworked back in March, so the label was catching up to a bill that had shifted a quarter earlier.

Here is where things stand today. The tool is one IDE with five ways to pay: a Free tier, Pro at $20 a month, a Max tier at $200, a Teams plan, and Enterprise on custom pricing. The rebrand was cosmetic in the way that matters to your wallet, because plans, extensions, and billing carried over untouched. What you pay for did not change on June 2. What it is called did.

What actually happened: the Cognition–Windsurf merger timeline

To understand the price, follow the money through the summer of 2025. Windsurf was a hot property, with roughly $82 million in annual recurring revenue and more than 350 enterprise customers, and three of the biggest names in AI wanted it.

• July 2025: OpenAI's reported $3 billion offer for Windsurf expired without a deal.
• Days later: Google paid around $2.4 billion in a reverse-acquihire, licensing Windsurf's technology and hiring CEO Varun Mohan, co-founder Douglas Chen, and research leaders into DeepMind.
• July 14, 2025: Cognition, the company behind the Devin coding agent, acquired what remained — the IP, the product, the trademark, and the rest of the team — for a reported $250 million.
• March 19, 2026: the pricing overhaul that retired credits and added the Max tier.
• June 2, 2026: the over-the-air rebrand from Windsurf to Devin Desktop.

That sequence — an expired megadeal, a talent raid, then a cut-price purchase of the leftovers — is the backdrop to everything in your billing settings. Google's move stripped out the founders and core researchers, and the broader agentic Gemini era Google unveiled at I/O 2026 shows where that talent went and why the labs were willing to spend billions on it. Cognition bought a product without its original leadership and had to make it pay for itself.

News that Cognition acquires Windsurf in July 2025 folded a popular IDE into a company that already ran an autonomous agent business. The two sides needed to speak the same commercial language, and that alignment is why the credit meter eventually had to go.

One more casualty is worth noting. Cascade, the original in-editor agent, reached end of life on July 1, 2026, replaced by Devin Local as the default local agent, a change documented alongside the model updates in this Cascade EOL and SWE-1.6 review. If your muscle memory still reaches for Cascade, that is another reason the tool feels unfamiliar.

Why did Cognition kill the Windsurf credit system?

Windsurf's old system billed in credits, and credits had a design flaw that anyone who used the tool heavily felt in their gut. A one-character autocomplete and a sprawling multi-file refactor could draw down the same balance. Users started rationing. They crammed several requests into one prompt to save credits, second-guessed whether a task was worth spending on, and watched a meter instead of writing code. The community called it credit anxiety, and it was real.

The Windsurf 2026 pricing overhaul on March 19, 2026 replaced that meter with daily and weekly quotas. Instead of paying per action, you get an allowance that refreshes on a schedule; go past it and overage bills at straight API rates rather than a marked-up credit multiplier. Tab autocomplete stayed unlimited on every plan, which quietly removed the most common trigger for credit panic.

The logic is worth stating plainly, because it drives the rest of the pricing. Quotas tie what you pay to compute you actually consume over time, not to a per-click abstraction. That shift from metering every action to metering sustained load, as the quotas replace credits analysis frames it, is exactly what a company selling flat $20 subscriptions needs. It also raised Pro from $15 to $20 in the same move, which matters once Cursor enters the picture.

How SWE-1.6 and the Agent Command Center unlocked $20 and $200

A flat $20 plan only works if your marginal cost per user stays low. For most AI coding tools, that cost is the API bill they pay to Anthropic or OpenAI on every completion. Cognition sidesteps a large chunk of it by running its own models.

SWE-1.5 is Cognition's frontier-size proprietary coding model, and its successor SWE-1.6 is post-trained on the same base. The SWE-1.5 announcement put the model on Cerebras hardware at up to roughly 950 tokens per second, about 13 times faster than Sonnet 4.5. Speed is not just a comfort feature here; it is the cost story.

The numbers that matter for a subscription line up cleanly. SWE-1.6 scores about 50.4% on SWE-Bench Pro, up from 40.1%, and it reaches its answers in roughly 40% fewer turns. Fewer turns means fewer tokens, and fewer tokens means lower cost per completed task. Serving those tokens on SWE-1.6 on Cerebras instead of a metered third-party API is what makes $20 a month sustainable rather than a loss leader.

The $200 Max tier answers a different question. Cognition's Agent Command Center lets power users run many agents in parallel, and parallel agents burn through a quota fast because each one is doing real work at the same moment. A single developer supervising one assistant fits comfortably inside Pro. A developer orchestrating a swarm does not. Max exists for that workload, and it belongs to the same shift toward MCP-first connectors running work through agents that is reshaping how these tools talk to each other.

Devin Desktop vs Devin Cloud vs Cursor: which pricing model fits your work?

Two lines run through this pricing, and confusing them is the most expensive mistake you can make. Devin Desktop is synchronous: you sit in the IDE, the agent works while you watch, and you pay a flat monthly subscription. Devin Cloud is asynchronous: you hand off a task, an autonomous agent runs it without you, and the meter runs in Agent Compute Units.

The pricing tiers at a glance

Here is every Desktop tier side by side, with the Cloud billing model called out separately so the two never blur together.

The row that trips people up is the last one. Everything above Enterprise is a subscription that covers you sitting at the keyboard. Devin Cloud work is billed on top, in ACUs, where roughly one ACU equals about 15 minutes of active autonomous execution. A flat subscription cannot absorb hours of unattended agent runtime, which is why autonomous work carries its own meter.

How it stacks up against Cursor

For a long time Windsurf's pitch against Cursor was simple: it cost $15 to Cursor's $20. The March increase erased that edge. Both now sit at $20 a month for their Pro plans, so the decision moves off price and onto everything else — model behavior, agent quality, and how each tool handles memory across a session, which the state of AI agent memory in 2026 breaks down across the major coding assistants.

The Cloud side of Devin invites a different comparison. Autonomous agents that run unattended are a young, failure-prone category, and understanding why they still break, covered in the state of real-time AI agents in 2026, matters more than the per-ACU rate when you are deciding whether to trust one with a backlog task.

ACU billing itself is not arbitrary either. Charging per unit of execution time reflects the underlying economics of running an agent, the same logic laid out in the 2026 agentic architecture reference framework. You pay for compute consumed, not for a seat that sits idle between runs.

The verdict: which Devin Desktop tier should you pick?

Match the tier to how you actually work, not to the biggest number you can justify on an expense report.

If you code casually, or you are still deciding whether an AI IDE belongs in your workflow, start on Free. Unlimited Tab autocomplete alone makes it a genuinely useful daily driver, and you will only bump the quota ceiling if you lean on the agent hard.

For the working developer — the person in the editor every day, shipping real features with an assistant at their side — Pro at $20 is the answer, and it lines up cleanly against Cursor at the same price. This is the tier most readers of this article should pick.

Max at $200 is not a status upgrade; it is a workload-specific tool. Pick it only if you genuinely run agents in parallel through the Agent Command Center and watch a Pro quota evaporate by mid-morning. If you are not orchestrating multiple agents at once, Max is money spent on headroom you will not touch.

Enterprise, with Devin Cloud ACUs, serves a different job entirely: autonomous, unattended work against a backlog, where you dispatch tasks and collect results later. If your question is how to hand off hours of work and walk away, you are shopping for ACUs, not a Desktop seat.

Key takeaways

• Devin Desktop is the renamed Windsurf IDE, owned by Cognition since the July 14, 2025 acquisition; the June 2, 2026 rebrand left pricing untouched.
• The March 19, 2026 overhaul retired credits for daily and weekly quotas, raised Pro from $15 to $20, and added the $200 Max tier.
• Proprietary SWE-1.5 and SWE-1.6 models served on Cerebras cut Cognition's API costs, which is what makes flat $20 pricing viable.
• Desktop subscriptions ($20 and $200) are synchronous; Devin Cloud autonomous work bills separately in Agent Compute Units, about 15 minutes each.
• Most developers want Pro; Max is for parallel-agent power users; Enterprise ACUs are for fire-and-forget autonomous runs.

FAQ

Is Windsurf still called Windsurf?

No. Cognition rebranded Windsurf to Devin Desktop on June 2, 2026 through an automatic over-the-air update. Existing installations updated themselves, and every plan, extension, and setting carried over — only the name on the window changed.

How much does Devin Desktop cost in 2026?

There are five options: a Free tier, Pro at $20 a month, Max at $200 a month, a Teams plan, and Enterprise on custom pricing. Tab autocomplete is unlimited on every one of them, so basic completion never counts against your quota.

Why did Cognition get rid of Windsurf credits?

Credits charged a complex agent task the same as a simple autocomplete, which bred credit anxiety and prompt-cramming. The March 19, 2026 switch to daily and weekly quotas, with overage billed at API rates, ties what you pay to the compute you actually use.

What's the difference between Devin Desktop and Devin Cloud?

Devin Desktop is the synchronous, human-in-the-loop IDE you drive yourself, priced at $20 or $200 a month. Devin Cloud runs autonomous agents without you and bills in Agent Compute Units, where one ACU is roughly 15 minutes of active execution. Desktop is you plus an assistant; Cloud is a machine working alone.

What is SWE-1.6?

SWE-1.6 is Cognition's proprietary fast coding model, the successor to SWE-1.5, served on Cerebras hardware at up to roughly 950 tokens per second. Its efficiency, reaching strong SWE-Bench Pro scores in about 40% fewer turns, is the technical reason flat $20 pricing holds together.

Start by naming the work in front of you, then let the tier follow: Free to try it, Pro for daily hands-on coding, Max for running agents in parallel, and Enterprise ACUs when you need tasks finished while you are away — and revisit the choice the next time your workflow shifts.