You're paying $20 a month for an AI coding tool. Maybe it's Cursor. Maybe it's Claude Pro to run Claude Code. Same sticker price. Totally different product inside the box.

And that difference decides who survives the next two years.

The question nobody asks

When you hand over $20, what are you actually buying? An interface? A model? A habit? The answer matters — because one of these companies is building on sand.

Cursor's bet: the IDE is the product

Cursor sells you an editor. You pay for Tab-autocomplete, for a polished UI, for the warm feeling of everything being in one place. The AI model inside? That's a cost center. Cursor buys tokens from Anthropic, OpenAI, Google — and resells them to you with a markup.

A monthly subscription where someone else's brain does the work. Pretty interface + deep integration + lock-in = recurring revenue. The oldest trick in SaaS.

The problem: Cursor sells a wrapper. And wrappers rot. VS Code — the free editor Cursor literally forked — already ships AI features through GitHub Copilot. When that reaches "good enough," who's still paying for the fork?

Nobody. That's who.

Anthropic's bet: the model is the product

Anthropic makes money on the API — the pipeline where developers send text to Claude and pay per token (the word-chunks AI processes, roughly ¾ of an English word). Claude Code isn't the product. It's the dealer's first hit for free.

More people use Claude Code → more API calls → more revenue. Razor-and-blades, except the razor is a CLI that lives in your terminal and the blades are compute. Even the $20/month Pro plan is a loss leader designed to keep you in the ecosystem until you can't imagine leaving.

Here's the part that should make Cursor sweat: Claude Code is an open client. Anthropic deliberately lets you plug in other models. Why would they do that? Because they know what happens next. You try Llama through OpenRouter, compare it to Claude Opus 4.6, realize the gap — and crawl back. That's not generosity. That's a company that knows its product is better and wants you to verify it yourself.

We've seen this movie

Netscape sold a browser in the 1990s. Microsoft made Internet Explorer free. Netscape died. Not because IE was better — but because "free" always wins when the product becomes a commodity.

Cursor is the Netscape of the AI-IDE moment. Excellent product. First to market. Loyal fans. Dead company walking the moment VS Code + Copilot reaches "good enough" — and "good enough" is Microsoft's entire business strategy.

Windsurf at least hedged by shipping their own models (SWE-1, SWE-1.5) in early 2026. That gives them control over unit economics — the math of cost-per-user versus revenue-per-user. Cursor depends entirely on other companies' models. That's not a business strategy. That's a prayer.

And those price tiers scream desperation. As of March 2026, Cursor Pro+ costs $60/month. Ultra is $200/month. When a product triples its price to $200, it's not "we deliver incredible value." It's "we're hemorrhaging money buying tokens from competitors and need you to cover the losses." The math doesn't lie, even when the marketing does.

The number that matters

According to Anthropic's benchmarks from February 2026, Claude Code uses 5.5x fewer tokens than competitors for equivalent tasks. You can debate the methodology. You can argue benchmarks are marketing. But even if the real number is 3x, that's a fundamentally different cost structure. Cursor can't match this because they don't control the models they run on. They're optimizing the menu while someone else sets the food prices.

In fairness

Cursor has UX. The VS Code extension ecosystem carries over. Developers who want visual feedback instead of a blinking cursor in a terminal — they exist, and there are millions of them.

Cursor could survive as a "premium IDE experience" the way JetBrains lives alongside free VS Code. But JetBrains builds their own compilers and debuggers. Cursor builds a skin on top of someone else's intelligence. That's a thinner moat than they'd like to admit.

What this means for you

The practical math, as of March 30, 2026:

  • Claude Code + local/free models via OpenRouter = $0. Handles routine tasks. No API key needed for free-tier models.
  • Claude Code + Pro = $20/month. Best value per dollar. You're buying the model directly.
  • Cursor Free = $0. Fine if you need an IDE with AI sprinkled in.
  • Cursor Pro = $20/month. Worth it only if the visual IDE measurably makes you faster — not "feels nicer," but measurably.

Don't pay for a wrapper when you can work with the source directly. If the wrapper genuinely makes you more productive, fine — that's your call. But understand what you're buying: a middleman's markup on someone else's intelligence, with all the risk that implies when the middleman's margins get squeezed.

In 2026, your money goes further on models than on interfaces to them. The interface is the part that gets commoditized. The brain is the part that doesn't.

→ Cursor Pricing → Claude Code Docs → GitHub Copilot → Anthropic SWE-bench Results