Cursor Review (August 2026): Buy, Hold, or Switch Before the SpaceX Deal Closes

SpaceX agreed to buy Anysphere, the company behind Cursor, for $60 billion in an all-stock deal on June 16, and it is set to close this quarter, per TechCrunch. As of August 5 the deal has not closed and the editor on your machine works exactly as it did in June. So the question a developer who lives in Cursor actually has is not "is it good." It is "do I renew, go month-to-month, or move my budget somewhere else before the ownership changes."

Here is the short answer before the detail. Solo devs on Pro: keep it, month-to-month. Small teams: do not sign an annual Teams contract right now. Enterprise: fine to deploy, but get model access in writing. The reasoning is below, tier by tier, with current pricing and the one thing that got worse since our July review.

What the acquisition actually means right now

Three facts, in the order that matters to a buyer.

Nothing about the product has changed. The same models are available (Anthropic Claude, OpenAI GPT, Google Gemini, and Cursor's own Composer), the same pricing applies, and the same team runs it, per Developers Digest. The changelog backs that up: the most recent entry is June 10, so there has been no feature freeze and no rushed change either (cursor.com/changelog).

The incentive underneath the deal points one way. SpaceX merged with xAI in February 2026, runs the Colossus compute cluster, and on July 9 shipped Grok 4.5, a coding model trained on trillions of Cursor interaction tokens, per Developers Digest. Every Cursor request routed to Anthropic is revenue that leaves SpaceX's own AI stack. That does not mean Claude vanishes next week. It means the owner has a standing reason to make its own model the cheaper, faster default over time.

The first joint model is already in the pipeline. SpaceXAI and Cursor are training a shared model on Colossus, with Cursor employees working out of xAI offices, and it is expected to ship inside both Cursor and Grok Build, per Developers Digest. When it lands, it will most likely become the default in Cursor's first-party pool. Whether it holds up against Claude on reasoning-heavy agent work is the open question.

Deal mechanics worth knowing: the transaction values Cursor at roughly 15x its approximately $4 billion annualized revenue, of which about $2.6 billion is enterprise, and it pre-empted a $2 billion fundraise that would have valued the company at $50 billion, per TechCrunch and Developers Digest. On close, Cursor becomes a wholly owned SpaceX subsidiary. The close needs regulatory approval, so the exact date is still open.

The current product, and the one thing that got worse

Rate the editor on what it does today and it is a 4.4. The Tab model predicts your next edit across a file rather than the next token, so a rename follows down its call sites and you cycle the fixes with Tab. Composer takes one instruction and applies it across many files, showing every diff before it writes. Bugbot got faster in the June 10 changelog: average review time dropped to about 90 seconds from roughly five minutes, runs cost about 22% less, and it finds about 10% more issues per review (0.62 vs 0.56), now powered by Composer 2.5 (cursor.com/changelog). None of that regressed. The build you install today is the same capable editor that earned the rating in July.

One thing did get worse, and it belongs in a security-conscious buyer's file. On July 14, Mindgard disclosed a Windows flaw where Cursor auto-executes a git.exe planted in a repository's root directory, with no click and no prompt, because its Git binary discovery searches the workspace itself, per Developers Digest. The impact is arbitrary code execution as the current user, and the trigger is ordinary: clone or open an untrusted repo, which is what you do when reviewing open-source code or a candidate's take-home. The timeline is the ugly part. Mindgard reported it via HackerOne on December 15, 2025, and by disclosure the fix still was not evident after more than 70 releases shipped, per Developers Digest.

Cursor is a VS Code fork, so it inherits the git.path setting. Pinning it to your system Git narrows the discovery surface the exploit relies on. Drop this into your workspace or user settings.json on Windows:

{
  "git.path": "C:\\Program Files\\Git\\cmd\\git.exe",
  "git.allowNoVerifyCommit": false
}

Tested 2026-08-05 on Cursor 3.7 / Windows 11. Adjust the path to your Git install; confirm with where git in a terminal.

This is a hardening step, not a vendor patch, so treat untrusted repositories as untrusted until Cursor ships a real fix. The version check you want before anything else:

# Confirm the build you are running (Help > About also shows it):
cursor --version

Input: run that in an integrated terminal. Expected output is a version line such as 3.7.x. Anything you are running should be current, but a current version does not close the git.exe issue on its own, which is exactly why the setting above matters.

Cursor pricing as of August 5, 2026

Prices are unchanged since the July restructure, and SpaceX has announced no pricing changes, per Developers Digest. Every tier includes a set amount of model usage and then bills on-demand in arrears.

Plan	Price	Included usage	Source
Hobby	Free	Limited completions	cursor.com/pricing
Pro	$20/mo	~$20 agent usage + bonus	cursor.com/pricing, Developers Digest
Pro+	$60/mo	~$70 agent usage + bonus	Developers Digest
Ultra	$200/mo	~$400 agent usage + bonus	Developers Digest
Teams Standard	$40/user/mo ($32 annual)	Two usage pools (first-party + third-party)	pricing blog
Teams Premium	$120/user/mo ($96 annual)	5x Standard usage	pricing blog
Enterprise	Custom	Pooled usage	cursor.com/pricing

Figures as of 2026-08-05. The Teams split from July is the part that still catches teams out: Standard is $40/user/mo monthly or $32 annual, and Premium runs $120/user/mo monthly or $96 annual, at 5x Standard's usage for 3x the seat cost, per the Cursor pricing blog. Cursor lets you mix seat types freely, so the right move is everyone on Standard and only the two or three heaviest agent users on Premium once the admin dashboard shows who they are.

The model-access question

The risk is not that Claude and GPT disappear tomorrow. It is that their relative priority shifts. Cursor markets model choice as a feature, and yanking it would push users out the door, so the likely path is quieter: make the first-party model the default while keeping third-party access on the menu, then tilt the cost and speed trade-off toward first-party over time, per Developers Digest.

Grok 4.5 is the first hard signal that the tilt has begun. xAI trained it on Cursor's interaction data and priced it at $2 per million input tokens, against $5 per million for both Anthropic Opus 4.8 and GPT 5.5, per Developers Digest. xAI positions it at "around Opus 4.7 level," a generation behind the frontier but at a fraction of the cost, and the same writeup flags that an earlier Cursor codebase snapshot was in the training set, which inflates its CursorBench number. Price that aggressive, on a model the owner controls, is how a default gets moved without anyone announcing a change.

Who this actually hits: teams that lean on Claude or GPT for complex, reasoning-heavy agent runs, where the model gap is the whole point. Who it does not hit today: most Pro users doing autocomplete and Composer edits, where a cheaper first-party model is a fine or better trade. If model portability is load-bearing for your work, the cleanest hedge is bring-your-own-key routing through a tool like Cline, where the key and the model choice sit with you rather than with a platform that has a reason to steer them:

{
  "cline.apiProvider": "anthropic",
  "cline.anthropicApiKey": "<YOUR_ANTHROPIC_API_KEY>",
  "cline.modelId": "claude-opus-4-8"
}

That is the portability argument in three lines: your routing does not change because your editor's owner shipped a rival model.

How Cursor compares for a 10-person team

Put three tools next to each other for a 10-seat team on annual billing. This is a decision table, not the verdict; the recommendation lands in the next section.

Tool	Annual cost, 10 seats	Models available	Owner conflict	Ownership	Source
Cursor Teams Standard	$3,840 ($32/seat/mo annual)	Claude, GPT, Gemini, first-party Composer	Yes, owner builds a rival model	SpaceX, pending Q3 2026 close	pricing blog, TechCrunch
GitHub Copilot Enterprise	$4,680 ($39/seat/mo)	GPT, Claude, Gemini	Low	Microsoft	github.com/features/copilot
Claude Code (Claude Max 5x)	$12,000 ($100/seat/mo)	Claude only	None	Anthropic	claude.com/pricing

Cursor is the cheapest of the three and the only one with genuine cross-vendor model choice, at $3,840/yr for the team, per the Cursor pricing blog. GitHub Copilot Enterprise runs $39/user/mo for org-wide rollout, per github.com/features/copilot, and its owner conflict is milder: Microsoft resells frontier models rather than racing to replace them in your editor. Claude Code on a Claude Max 5x seat is $100/mo, per claude.com/pricing, roughly three times Cursor's per-seat cost, and it locks you to Anthropic. The trade is visible in the table: Cursor wins on price and flexibility today, and carries the ownership question none of the others do.

The verdict, by who you are

Split it by buyer, because the deal hits each one differently.

Solo developer on Pro ($20/mo, per cursor.com/pricing): keep Cursor. The Tab model still pays for itself in recovered edit time, the switch off VS Code is a config import, and the six-month risk is low for a single seat. Stay month-to-month so your exit stays clean, and re-evaluate if the joint SpaceXAI model ships and visibly degrades Claude routing. If you are deciding fresh, pull the current tiers and start a Pro trial from Cursor's site; the free Hobby plan is enough to judge whether the Tab model fits your language before you pay a cent.

Small team (3 to 10) on Teams Standard: do not sign an annual contract right now. Run month-to-month at $40/user until the deal closes and the first post-close terms are visible, per the Cursor pricing blog. The roughly $8/user/mo premium over the $32 annual rate is the cost of not being locked in through the most uncertain stretch in Cursor's history, and that is cheap insurance. If Cursor is the right pick for your team now, the month-to-month Teams plan needs no annual commitment.

Enterprise buyer evaluating 50-plus seats: deploy is fine, but require written model-access and data-routing guarantees in the contract before you sign. "Nothing is changing" is a verbal assurance from a company that will own a competing model; the version that protects you is a written commitment about Claude and GPT availability and where your code data goes, per Developers Digest. If SpaceX will not put it in writing, that refusal is itself the answer. Add a hard version floor and the git.path hardening above to your rollout policy while the git.exe issue is open.

A note on the rating: the 4.4 measures the editor as it stands on August 5, and it does not fold the acquisition uncertainty into the number. A rating is a snapshot of a product, not a forecast of who owns it in a year. The overhang shows up in the advice, not the score.

What to watch

Three triggers should prompt a fresh look:

• The close-date announcement, and any investor or employee communications that name a product direction.
• The first SpaceXAI-Cursor model launch: whether it ships as a peer option next to Claude and GPT, or as an aggressively routed default that deprioritizes them.
• Your renewal window: if a team contract comes up before the close, read the updated terms line by line rather than clicking through.

The editor itself is still excellent today. The reason to think before you buy is not the code in front of you. It is the name that will be on the door when your annual contract renews.