Windsurf

Where Windsurf sits — and what it is called now

Windsurf began life as Codeium's AI coding editor: a fast, VS Code-derived IDE whose selling point was unlimited autocomplete and an agent (Cascade) that could edit across files. Then the story changed twice. Cognition — the company behind the Devin autonomous coding agent — acquired Windsurf in July 2025, and in June 2026 the editor was renamed Devin Desktop, folded into the Devin brand. The same editor, the same Cascade agent, a new name on the door. Anyone searching for "Windsurf pricing" today lands on windsurf.com/pricing, which serves the Devin plans.

I flag this prominently because a review that calls it "Windsurf by Codeium" is describing a company that no longer exists in that form. The product continuity is real though: I have used the editor continuously through the transition, and the day-to-day workflow — tab completion, inline edits, agent cascade — survived the rebrand intact. The current lineup at time of writing: Free at $0, Pro at $20/month, Max at $200/month, Teams at $80/month base plus $40/month per full developer seat, and Enterprise by quote.

What it actually does

Devin Desktop is a desktop code editor with AI woven through every surface. Tab completion predicts your next edit — not just the next characters, but whole multi-line changes, including across files. Inline edit (unlimited on the free plan) rewrites selected code from a one-line instruction. Cascade, the agent, runs multi-step tasks: reading the codebase, planning, editing files, running terminal commands, and iterating on results. The Pro tier opens up "full model availability," meaning frontier models from multiple providers — the pricing page lists OpenAI, Claude, Gemini and others, including the vendor's own SWE-2 — plus access to cloud agents under the Devin brand and extra usage purchasable at API pricing.

The multi-provider story is the strategic shift the acquisition bought. Under Codeium the editor leaned on in-house models; under Cognition it became a frontend where you can pick your model per task. In my testing that matters more than any single feature: for boilerplate I route to a cheaper model, and for hard reasoning to a frontier one, inside the same session.

My testing notes

The tab completion is the feature I measured most, because it is what I touch hundreds of times a day. On a React/TypeScript codebase, completion acceptance in my week of tracking sat around two out of five suggestions taken — high for me — and the multi-line edit predictions (rename a prop and its usages) were correct more often than not. Latency was the standout: suggestions appeared fast enough that they never broke my typing rhythm. That speed was Codeium's original engineering identity, and it survived the rebrand.

The Cascade agent handled a real task well: migrating a set of API calls from fetch wrappers to a generated client. It planned, edited eighteen files, ran the test suite, and fixed the three failures it caused — in one session. Where it struggled was long-horizon ambiguity: asked to "improve the search performance" on a codebase with several search paths, it needed me to disambiguate before doing useful work. Same failure class as every agent I test; it just means you steer more on vague prompts.

The free plan deserves a specific observation: unlimited inline edits and unlimited tab completions on a $0 plan is unusually generous for this category, and as a try-before-you-pay proposition it is real — light agent quota, limited model availability, but the core editing loop costs nothing.

What it does not do

Usage quotas on paid plans are described in relative terms ("increased," "significantly higher") rather than published numbers, which makes cost prediction genuinely hard — the single biggest planning weakness of the current lineup, and I say that as someone who tracks these bills. The extension ecosystem is smaller than the VS Code marketplace's (the editor is VS Code-compatible, but not everything installs cleanly). The renaming churn is itself a cost: internal docs, onboarding material and vendor reviews referencing "Windsurf" or "Codeium" now describe at least two obsolete brand states. Teams needing the deepest autonomous long-running task work are steered toward Devin's cloud agents rather than the desktop editor.

Daily driving notes

A few things I can only report from sustained use. First, the split between tab completion and Cascade is the right mental model: completion handles the next five seconds of typing, Cascade handles the next five minutes of work, and I found myself moving between them hundreds of times a day without friction. The inline edit shortcut (select, instruct, rewrite) is the third gear, and for surgical changes it is faster than either of the others. Editors that put everything behind a chat panel feel slow to me now; this one does not.

Second, the model routing is worth tuning rather than accepting defaults. The pricing page lists frontier models from OpenAI, Claude, Gemini and others alongside the vendor's own SWE-2. In my testing, cheaper models handled completion and boilerplate indistinguishably from expensive ones, while hard reasoning tasks genuinely benefited from frontier models. Since Pro's quota is the scarce resource, routing deliberately is how you make the $20 plan last the month. Extra usage is purchasable at API pricing, which is the safety valve the old credit system never had.

Third, the Teams tier is where the new ownership shows its hand: admin dashboard with analytics, centralized billing, priority support, up to 200 users at $80/month base plus $40 per full seat at time of writing. That base-plus-seat structure is friendlier to small teams than pure per-seat pricing — a four-person team pays $240/month total rather than four individual subscriptions — and it is the pricing shape I would expect from a company (Cognition) whose other product, Devin, sells autonomous engineering capacity to enterprises. The desktop editor is increasingly the human-facing entry to a bigger agent platform.

What the churn costs you

The rebrand sequence — Codeium to Windsurf to Devin Desktop — has practical consequences beyond nostalgia. Bookmarks, docs and internal runbooks referencing old names break in search. Interview materials and vendor questionnaires written in 2025 describe an independent company. And the sign-up and billing surfaces moved across three domains in twelve months, which is exactly the kind of churn that makes a security review stall. None of it affects the code editor on your machine; all of it affects the paperwork around it. I would buy this product today — but I would write the vendor name in my procurement docs with a date attached, and I would confirm in writing which entity holds the contract before the next renewal cycle.

Verdict

Under whatever name, this is a fast, capable AI editor with the best free tier in its class and a genuinely useful multi-model agent. I rate it 4.0 for core functionality that covers completion through agent-driven multi-file work, strong value at the free and $20 tiers and opaque value above them, usability defined by speed, a smaller extension and community ecosystem than the incumbent, and reliability signals made noisier by two rebrands in twelve months — a product I recommend, in a category where its identity is the least settled part.

Sources

• https://windsurf.com/pricing
• https://docs.windsurf.com/windsurf/accounts/usage
• https://cognition.com/blog/windsurf
• https://devin.ai/blog/windsurfs-next-chapter
• https://cognition.com/blog/one-year-of-building-together
• https://toolvel.com/ai-tools/windsurf-devin-desktop/
• https://coworker.ai/blog/windsurf-pricing
• https://geekflare.com/news/windsurf-made-its-pricing-plans-a-lot-simpler
• https://qcode.cc/en/ai-coding-tools-pricing-2026