Decisions

Cursor's Cheapest Seat Runs a Different Model

Cursor launched a 649 rupee India tier on 28 July. It keeps the two models Cursor does not buy and drops the two it pays for. That is a bill of materials, not a discount.

DecisionsBy Servola Tech Desk2026-07-285 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Cursor's Cheapest Seat Runs a Different Model

Key takeaways

Cursor launched Start in India on 28 July at 649 rupees a month, tax included, against 20 dollars a month for Pro. The cheaper seat keeps Cursor's own Composer model and Grok 4.5 and removes the frontier models from OpenAI and Anthropic.
Read the two lists side by side before you read the price. The models that stay are the ones Cursor does not buy from a third party. The models that go are the ones it pays another lab for. The discount is a bill of materials, not a currency adjustment.
SpaceX is acquiring Cursor's parent Anysphere for about 60 billion dollars in stock, expected to close this quarter. When it does, the cheapest seat in the tool will route by default to a model owned by the same group.
Assume the tier structure travels. A cheaper European seat is a reasonable thing to want, and the thing it will cost you is not features, it is which lab writes your code.
Before your next renewal, get the per-tier model list into the contract in writing, because the tier name has stopped describing what you are buying.

The price Cursor took to India

On Monday, Simon Green, who runs Asia-Pacific and Japan for Cursor, explained the reasoning behind a number the company published the next morning. The company had, he said, an opportunity to right-size the commercial model and drive scale. The number is 649 rupees a month, tax included, roughly seven dollars, payable by UPI or card, and available only in India. Cursor calls the tier Start, and its own announcement gives a plain reason for it: users there had been asking for two things, pricing that reflects the local market and the ability to pay with UPI.

The market case is not in doubt. India is Cursor's third-largest, it has more than three million users there, that base has more than tripled in a year, and the company says it holds the highest concentration of power users anywhere. Everyone covering this led on the geography. The interesting part is on the other page of the announcement, in the list of what the cheaper seat contains.

Read the model list, not the price

Start includes Cursor's own Composer model and Grok 4.5. It also carries always-on cloud agents, the iOS app, plugins, Model Context Protocol servers, hooks and skills, and more agent requests than the free tier. That is a real product, not a crippled demo.

What it does not include is the part that matters. Start has no access to the frontier models from OpenAI and Anthropic, which the 20-dollar Pro tier does carry, and it drops Bugbot, Auto Mode, Automations and the Cursor SDK.

Put the two lists next to each other and the pattern is immediate. Everything that stays is something Cursor does not buy from a third party. Everything that goes is something Cursor pays another company for on every call. This is not a currency decision and it is not generosity. It is a bill of materials with a country name written on the front of it.

The second model in that list is about to be in-house

Grok comes from xAI. SpaceX agreed in June to acquire Anysphere, Cursor's parent, for about 60 billion dollars in an all-stock deal expected to close this quarter, subject to approval. Both xAI and SpaceX sit in the same ownership orbit.

So the entry tier of one of the most widely used AI coding tools in the world is being defined, weeks before an acquisition closes, around a model belonging to the acquiring group and around a model Cursor built itself. Nobody has to allege a plan for that to matter. The commercial logic gets there on its own: the tier with the lowest price is the tier with the lowest cost of goods, and after the deal completes, that will mean the group's own model.

This is vertical integration arriving in developer tooling, and it arrives wearing the clothes of a favour. That is the form it usually takes. The cheapest option is rarely presented as a narrowing of choice, but a default that costs nothing is the most durable default there is.

Assume the tier travels

Purchasing-power pricing is ordinary in software and it is not the story here. What is worth planning around is that Cursor has now demonstrated a shape: a cheaper tier that is cheaper because the expensive models are missing from it. Once a company has built that shape and priced it, it does not stay in one country.

The European version will not be 649 rupees. It may well be a mid-priced seat pitched at teams under budget pressure, and it will be attractive, because most engineering work does not need a frontier model and paying for one on every autocomplete has always been poor value. The point is not to refuse it. The point is to know precisely what you are agreeing to when your finance director asks why the tooling line cannot come down, and the honest answer turns out to be that it can, at the price of which lab writes your code.

One line to add before you renew

The practical response is a single clause and a single audit. The clause: the contract names which models each tier grants access to, and requires notice before that list changes. Today most agreements name a tier and a seat count, which as of Monday no longer describes the product. A vendor can change the model behind a tier without touching the price, the name or the feature list, and nothing in a standard subscription stops it.

The audit is smaller. Ask your engineering leads which model their seats are actually calling, by tier, this week. In most companies nobody currently knows, because until recently the answer was implied by what you paid. It is not implied any more.

Then decide deliberately. A cheaper seat running a capable non-frontier model is a perfectly good choice for a great deal of work, and for some teams it will be the right one. It stops being a good choice when it is the result of a renewal nobody read closely, on a tier whose contents changed after signature.

Frequently asked questions

Is Cursor Start available outside India?

No. The tier is India-only, billed in rupees and payable by UPI or card. Cursor has not announced equivalent regional tiers in Europe or elsewhere, which is why the useful posture is preparation rather than reaction.

Does the cheaper tier still do real work?

Yes. It carries Cursor's own Composer model and Grok 4.5, always-on cloud agents, the iOS app, plugin and Model Context Protocol support, hooks and skills, with higher usage limits than the free tier. The gap against Pro is access to the frontier models from OpenAI and Anthropic, plus Bugbot, Auto Mode, Automations and the SDK.

Why does the SpaceX acquisition matter to a European buyer?

Because it changes who owns the default. Grok comes from xAI, and SpaceX expects to close its acquisition of Cursor's parent this quarter. Once it does, the lowest-priced tier of a widely used coding tool routes to a model inside the acquiring group, and defaults set at the cheapest tier tend to become the most common configuration in a company.

What should we actually change this quarter?

Two things. Put the per-tier model list into the contract with a notice requirement before it changes, and find out which model each of your existing seats is calling today. Both are small pieces of work, and both stop being possible to do calmly once a renewal is already on the table.

The lesson here is older than this product and it applies well beyond one coding tool. When a vendor offers you the same thing for less, the first question is what is no longer in it, and the answer is increasingly a component you cannot see from the invoice. Cursor has been unusually clear about the swap, publishing the model list next to the price where anyone can read it. The companies that get caught by this will not be the ones that took the cheaper tier. They will be the ones that never checked which model came with it.

CursorAnysphereSpaceXGrokAI coding toolssoftware procurementregional pricing

Rotate Every Secret Grok Build Ever Saw

A wire-level analysis found xAI's Grok Build CLI uploaded entire repositories, including a tracked .env file with live keys, to an xAI storage bucket - and the privacy toggle did not stop it. What every team that ran it owes its customers now.

6 min read

Artificial Intelligence

Grok 4.5 Loses the Benchmarks and Wins the Bill

xAI launched Grok 4.5 at 2 and 6 dollars per million tokens, a third of Opus, and it finishes tasks with 4.2 times fewer output tokens. It loses the benchmarks and wins the bill.

3 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.