Cursor has become the fastest-growing product in the history of developer tools by one closely watched metric: annualized revenue. Anysphere, the company behind Cursor, reportedly crossed $2 billion in annualized recurring revenue in February 2026, doubling from $1 billion just three months earlier, with more than one million paying customers and roughly one million daily active users. Those are extraordinary numbers for a code editor. But revenue growth answers a business question, not a product question. The question developers actually care about is simpler: does Cursor make you faster at writing software? The answer, according to the best evidence available, is “it’s complicated” — and the complications are worth understanding before you pay for a subscription.
What Cursor Actually Is
Cursor is a fork of Visual Studio Code built by Anysphere, a San Francisco startup founded in 2022 by Michael Truell, Sualeh Asif, Arvid Lunnemark, and Aman Sanger, with early seed funding from the OpenAI Startup Fund. Rather than bolting AI onto an existing editor as an extension, Cursor rebuilds the editing experience around AI: predictive “Tab” completion that anticipates your next edit, a chat interface with full codebase awareness, a multi-file editing mode called Composer, and an autonomous “Agent” mode that can plan, write, run, and debug code across a session with minimal supervision. It supports multiple frontier models — Anthropic’s Claude family, OpenAI’s GPT line, and Google’s Gemini — inside a single interface, along with native support for the Model Context Protocol (MCP) that lets it connect to external tools and data sources.
In practice, this makes Cursor less like an autocomplete plugin and more like a junior collaborator sitting inside your editor. That framing is also where most of the controversy about its actual value begins.
Pricing: Simple on the Surface, Complicated Underneath
Cursor’s pricing looks straightforward at first glance: a free Hobby tier, Pro at $20 per month, and Business at $40 per user per month. Dig one level deeper and the structure gets more complex — current published tiers span Hobby, Pro, Pro+, Ultra, Teams, and Enterprise, with Ultra reaching roughly $200 per month for the heaviest individual users.
The bigger shift happened in June 2025, when Cursor replaced flat per-user request limits with usage-based billing tied to the underlying model’s compute cost. Heavier models, longer context windows, and “Max Mode” reasoning now consume a larger share of a user’s included allowance, while lighter autocomplete-style usage barely registers. The logic — align price with actual compute — is defensible. The execution was not well received: the rollout was widely described as rushed and poorly communicated, and it remains a recurring source of frustration in developer communities. A developer who casually turns on Max Mode with a frontier reasoning model can watch a predictable $20 monthly bill balloon toward $120 without much warning. Anyone evaluating Cursor should budget for this variability rather than assuming the advertised tier price is the final cost.
The Real Question: Does It Make You Faster?
This is where Cursor’s story gets genuinely interesting, because it is one of the few AI products to have been the subject of a rigorous, independent randomized controlled trial — the gold-standard research method more commonly associated with medicine than software.
In July 2025, the nonprofit research organization METR (Model Evaluation & Threat Research) published a study titled “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity.” Researchers recruited 16 experienced open-source developers, each with an average of five years working on the specific repositories involved, and had them complete 246 real tasks — actual bug fixes, features, and refactors their projects needed. Each task was randomly assigned to either allow or disallow AI assistance. When AI was permitted, developers primarily used Cursor Pro paired with Claude 3.5 or 3.7 Sonnet.
The result surprised the researchers themselves, who said they had expected to find a speedup. Instead, developers who used AI tools took 19% longer to complete their tasks than those who worked without them. Even more striking was the perception gap: before starting, developers predicted AI would cut their completion time by 24%. Afterward, despite the measured slowdown, they still believed AI had made them roughly 20% faster. In other words, the subjective sense of productivity and the objectively measured outcome pointed in opposite directions — a nearly 40-percentage-point gap between belief and reality.
It would be a mistake to treat this as the final word, and to METR’s credit, the organization has continued tracking the question rather than resting on a single headline number. In February 2026, METR published a follow-up update acknowledging a serious methodological problem with its second attempt at the study: as AI tools became more embedded in developers’ daily workflows through late 2025, 30–50% of invited participants declined to complete tasks without AI access at all, even when offered $50 per hour to do so. This selection effect biased the available sample toward developers less representative of typical AI usage, making the follow-up data an unreliable signal on its own. Within that imperfect newer cohort — 57 developers across more than 800 tasks — the measured effect shifted to a 4% slowdown, but with a wide confidence interval spanning from a 15% slowdown to a 9% speedup, meaning the true effect could plausibly sit on either side of zero. METR’s own stated conclusion, based on this data plus qualitative feedback from participants, is that AI likely provides productivity benefits in early 2026 — a real reversal in direction from July 2025 — but the organization is explicit that the size of that benefit remains genuinely uncertain and is redesigning its methodology to measure it more reliably.
The honest synthesis: the crude “AI coding tools make you X% faster” claim that vendors lean on is not well supported by the best independent evidence available, in either direction, as a universal number. What the data does support is that outcomes vary enormously by task complexity, developer experience with the tool, and how the tool is used — junior developers and those working on well-scoped, familiar-pattern tasks tend to see more consistent gains than senior engineers working on complex, unfamiliar systems, who often spend significant extra time reviewing and correcting AI-generated suggestions.
Where Cursor Genuinely Excels
None of this means Cursor is a bad product — it means the productivity story is more nuanced than marketing suggests. Where Cursor consistently draws praise, including from reviewers who are otherwise skeptical of AI coding hype, is in specific, bounded use cases: scaffolding new projects, writing test suites, navigating unfamiliar codebases, handling repetitive refactors across multiple files via Composer, and reducing the friction of context-switching between documentation and code. Its multi-model access — being able to route a task to Claude, GPT, or Gemini depending on which handles it best — is a genuine convenience advantage over single-model competitors like GitHub Copilot.
The Agent mode, which can autonomously plan and execute multi-step coding tasks with a review step at the end, represents the frontier of what’s currently possible in this category, and it is the feature most tied to Cursor’s fastest-growing usage. But autonomy cuts both ways: it also means more code enters a codebase without a human writing every line, which raises downstream code review and maintainability questions that are separate from raw completion-time metrics.
How It Compares
Cursor’s most direct competitors are GitHub Copilot, priced at roughly $10 per month for individuals and $19 per user for Business, and Windsurf, which entered the market at $15 per month specifically to undercut Cursor after the June 2025 pricing backlash. Copilot remains the safer, cheaper, more predictable choice for developers who want strong autocomplete without an unfamiliar editor or a variable bill. Windsurf has positioned itself as the value alternative for developers who want Cursor-like capability without Cursor-like pricing volatility. Cursor’s advantage over both is depth: Composer, Agent mode, and multi-model flexibility go further than Copilot’s more conservative feature set, and its Fortune 500 enterprise adoption suggests the depth is paying off for teams with complex codebases.
Hidden Costs to Watch For
Beyond the headline subscription price, three factors tend to catch new users off guard. First, the credit system does not roll over unused capacity month to month on most tiers, so bursty usage patterns — a quiet week followed by a crunch week — can leave you paying overage fees even if your average monthly usage is modest. Second, model choice matters enormously to cost: routing every query to a frontier reasoning model in Max Mode burns credits far faster than using lighter, faster models for routine completions, and Cursor’s interface does not always make that trade-off obvious in the moment. Third, teams evaluating Business or Enterprise tiers should budget time for onboarding, since admin controls, SSO, and privacy-mode configuration add setup overhead that a solo Pro subscription does not require. None of these are dealbreakers, but they are the kind of detail that a quick glance at the pricing page will not surface.
The Verdict
Cursor is not a magic productivity multiplier, and anyone selling it as one is overstating the current evidence. It is, however, a genuinely capable, actively improving tool that different developers will get very different value from depending on what they build and how they work. For professional developers doing complex, multi-file engineering work — the use case Cursor is explicitly built for — the $20 monthly Pro tier is a reasonable bet, provided you keep an eye on credit consumption once Max Mode enters the picture. For developers doing lighter, simpler work, the free tier or a cheaper competitor likely delivers comparable value at a fraction of the cost. The most useful thing any prospective buyer can do is ignore both the vendor’s productivity claims and the loudest social-media enthusiasm, and instead treat Cursor the way METR does: as a tool whose real-world impact is still being measured, one careful study at a time.