Antigravity vs Cursor vs Claude Code: Free vs $200 [2026]

Three companies just placed very different bets on how developers write code in 2026. Google shipped an agent-first platform called Antigravity that hands off entire tasks to AI rather than waiting for keystrokes. Anysphere kept building Cursor into the AI-native editor that made autonomous coding a mainstream habit. Anthropic doubled down on Claude Code, a terminal-native agent with no graphical editor at all. Each one solves the same problem, turning a prompt into working software, in a genuinely different way, and picking the wrong one costs real money and real engineering time.

This comparison breaks down Google Antigravity, Cursor, and Claude Code on pricing, model support, agent behavior, and benchmark data pulled from vendor disclosures and independent reviewers. We also show where GitHub Copilot and Windsurf still fit, because most teams choosing an AI coding assistant right now are really choosing between five tools, not three.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

Meet the Contenders: Three Different Bets on AI-Assisted Coding

Before the spec sheets, it helps to know where each product came from. They were not built by the same kind of company, for the same kind of user, and that history still shows up in how each one behaves today.

Google Antigravity: The Agent-First Platform

Google announced Antigravity on November 18, 2025, alongside the Gemini 3 model family, positioning it as an answer to agentic tools like Cursor rather than a traditional code editor – the same month Claude Code itself moved from its February 2025 preview into full general availability, so both platforms effectively entered the market within weeks of each other. According to Google’s own developer blog, Antigravity treats the editor, the terminal, and the browser as one surface an agent can act across, planning a task, executing it, and verifying the result without a human approving every step. It launched as a free public preview for individuals, a deliberate move to seed adoption fast. At Google I/O 2026, the company expanded it into Antigravity 2.0, adding a standalone desktop app, a CLI, and an SDK, a move TechCrunch reported on May 19, 2026, and by October 2026 that 2.0 line had iterated forward to version 2.21.1. That 2.0 release left Antigravity even further behind on paper: Anthropic had already pushed Claude Code through v2.1.90 in April 2026, adding NO_FLICKER rendering and a 60% speed boost to its Write tool according to Apiyi Help, then to v2.1.101 on April 10, 2026 with /team-onboarding support and OS-level certificate authority trust, and by September 2026 the CLI itself had switched to spawning a native binary per Claudefa.st, though a higher version number alone says little about which platform’s agent actually performs better on a given task.

Cursor: The AI-Native Editor That Went Mainstream

Cursor comes from Anysphere, founded in 2022 by four MIT students: Michael Truell, Sualeh Asif, Aman Sanger, and Arvid Lunnemark. The editor itself launched in March 2023, back when GitHub Copilot had the AI coding market largely to itself. Anysphere’s funding history since then reads like a speed run: an Windsurf had raised about $243 million total, and the reported figures are the failed $3 billion OpenAI acquisition and the later $2.4 billion Google licensing-and-hiring deal.[1][12]9 billion. Five months later, a $2.3 billion Series D co-led by Accel and Coatue pushed that valuation to $29.3 billion, with TechCrunch confirming Nvidia and Google joined the cap table as new investors. CNBC’s coverage of the same round noted the pace, not just the size, as the standout detail. By April 2026, Anysphere was reportedly in talks for a further round near a $50 billion valuation, though that round had not closed as of this writing.

Claude Code: Anthropic’s Terminal-Native Agent

Claude Code takes the opposite design bet from both of the above. There is no dedicated window, no file tree, no click-to-accept diff view by default. It lives in the terminal, reads and edits files directly, runs tests, and reports back in text. Anthropic bundles it into paid Claude subscriptions rather than selling it as a standalone product, which means its price is really the price of a Claude plan, and as of September 2026 that access comes automatically with every Team plan’s standard seat, according to Anthropic’s own Help Center, not just its higher usage tiers. Anthropic’s own page now describes Claude Code for every seat, with self-serve seat management and up to $500 per month in pooled API credits, and on October 24, 2026 the company published “The making of Claude Code,” describing its evolution from an internal CLI into Anthropic’s coding agent. The entry-level Pro plan itself has held steady at $20 a month all the way through September 2026, a rare stretch of price stability in a category where competitors keep adjusting their tiers. That single decision shapes almost everything else about how teams evaluate it against Antigravity and Cursor.

Specs at a Glance: The Full Comparison Table

Here is how the three tools stack up on the specs that actually change a purchasing decision.

Spec	Google Antigravity	Cursor	Claude Code
Maker	Google DeepMind	Anysphere	Anthropic
First released	November 18, 2025	March 2023	2025 (CLI preview), expanded through 2026
Interface	Editor, desktop app, CLI, browser control	Standalone IDE (fork of VS Code)	Terminal / CLI only, no dedicated GUI
Built on	VS Code fork, per secondary reporting	VS Code fork	Native CLI, works alongside any editor
Free tier	Yes, Free plan at $0 with basic weekly rate limits	Yes, Free plan	No standalone free tier, requires a paid Claude plan
Entry paid price	Bundled into Google AI Pro (~$20/mo consumer plan)	$8/month (Plus)	$20/month (Pro, $17/mo billed annually)
Top individual tier	Google AI Ultra, up to $200/month	$200/month (Ultra)	$200/month (Max 20x)
AI models available	Gemini 4 Argon, Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5, GPT-OSS 120B	User-selectable Claude and GPT models, billed separately at API list prices	Anthropic’s own Claude model family only
Agent mode	Yes, core to the product (plan, execute, verify)	Yes, Composer multi-file agent mode	Yes, the entire product is agentic by design
Browser control	Yes, agents can drive a browser directly	No native browser control	No native browser control
MCP support	Yes	Yes	Yes
IDE ecosystem	Own editor plus CLI and desktop app	Own editor only	Terminal-first, pairs with any IDE
Codebase indexing	Yes, project-wide context	Yes, with a claimed 200K-token context window	Yes, reads the working directory directly

The single biggest structural difference is model choice. Antigravity is the only one of the three that hands you a menu spanning three different AI labs in one interface: a menu that began at launch with Google’s own Gemini 3 Pro, Anthropic’s Claude Sonnet 4.5, and an open-weight GPT-OSS option, according to Google’s launch post, and that Google’s own pages now show as Gemini 4 Argon, Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5, and GPT-OSS 120B. Cursor lets you pick between Claude and GPT models too, but stays inside a conventional editor shell. Claude Code has no model picker at all. You get Anthropic’s models, full stop, which is either a limitation or a feature depending on how much you trust one lab’s outputs. The models doing the reasoning today are Gemini 4 Argon at the top of Antigravity’s list and Claude Opus 5.5 inside Claude Code, a reminder that the real gap between these tools is model performance dressed up as interface design.

Pricing Compared: Free Tiers to $200 a Month

Pricing across all five tools converges on a strange coincidence: the top individual tier for Cursor, Claude Code, and Windsurf all land at or near $200 a month, while Antigravity’s core experience remains free during its public preview. Cursor’s version of that $200 tier changed on October 12, 2026, when, according to reports, it moved new customers to The X Subscription: Free, Plus at $8 a month, Premium at $30, Super at $100, and Ultra at $200, with the old $20 Pro plan now treated as a legacy plan for new customers. The new Ultra tier includes a shared Grok usage pool, X Premium benefits, and Grok Bot, while Claude, GPT, and Gemini models remain usable on Premium and above but are billed separately at API list prices through prepaid credits. Here is the full breakdown, including GitHub Copilot and Windsurf for context.

Plan tier	Antigravity	Cursor	Claude Code	GitHub Copilot	Windsurf
Free	Free (public preview)	Free	Not offered standalone	Free (2,000 completions + 50 chat/mo)	Free
Entry individual	~$20/mo via Google AI Pro	$8/mo (Plus)	$20/mo (Pro)	$10/mo (Pro)	$20/mo (Pro, since March 19, 2026)
Mid tier	N/A	$30/mo (Premium), $100/mo (Super)	$100/mo (Max 5x)	$39/mo (Pro+)	N/A
Top individual tier	~$200/mo via Google AI Ultra	$200/mo (Ultra)	$200/mo (Max 20x)	$39/user/mo (Enterprise)	Custom (Enterprise)
Team / Business	Not broken out separately	$40/user/mo	Team plan, custom seat pricing	$19/user/mo (Business)	$40/seat/mo (Teams)

A few things stand out once the numbers are next to each other. GitHub Copilot’s $10 Pro plan remains one of the cheapest paid entry points in the category, undercut only by Cursor’s new $8 Plus tier, which is exactly why Microsoft keeps winning the budget-conscious segment. Windsurf’s March 2026 repricing quietly erased its old advantage: Pro moved from $15 to $20 a month and Teams from $30 to $40 a seat, per Cognition’s own pricing update, landing it on the price of Cursor’s now-legacy $20 Pro plan instead of undercutting it. Claude Code’s Max 5x and Max 20x tiers exist specifically for developers who burn through tokens fast on agentic, multi-file tasks, especially since /loop scheduled tasks reached general availability in March 2026, letting those plans run cron-style recurring jobs unattended, per Apiyi Help, and Anthropic doubled the usage limits on those paid plans on May 6, 2026, a sign that the original caps were biting harder than the company expected.

Antigravity’s pricing is the hardest to pin down cleanly, and that is worth flagging rather than glossing over. The core agent experience is free during public preview, but heavier usage draws on the broader Google AI subscription bundle, and reported figures for that bundle do not fully agree with each other. MorphLLM pegged Google AI Pro at exactly $19.99 a month in March 2026, and AntigravityLab’s breakdown that same month put AI Ultra’s 5x Pro limits at $100 a month. By August 2026, Softr.io was still listing Google AI Ultra at that same $100-a-month figure for Antigravity access, alongside Claude Code’s Max tiers running $100 to $200 a month for comparison, while Sabaoon.dev reported a considerably wider spread back in April 2026, pricing Antigravity Ultra at $249.99 a month against Claude Code Max at $100 a month, a $149.99 gap. That gap closed by May 2026, when Google formalized a dedicated $100-a-month Ultra plan offering 5x Pro capacity and brought its top Ultra tier down from roughly $250 to $200 a month, landing it right next to Claude Code’s own $200 Max ceiling. That is a subscription to Google’s whole AI stack, not a line-item price for Antigravity alone, so treat any single number here as an estimate rather than an official per-seat rate.

Which AI Models Power Each Tool

The model question matters more than most feature lists suggest, because the editor or agent shell is really just the interface sitting on top of a model doing the actual reasoning. Get the model wrong and no amount of tooling polish fixes it.

• Google Antigravity lists Gemini 4 Argon and Gemini 3.8 Flash alongside Claude Sonnet 5.5, Claude Opus 5.5 and an open-weight GPT-OSS 120B model, giving teams a genuine multi-vendor choice inside a single agent platform.
• Cursor lets developers manually select from premium Claude and GPT models depending on the task, billed through Cursor’s own credit system rather than a direct API pass-through.
• Claude Code runs exclusively on Anthropic’s Claude model family. Its newest flagship is Claude Opus 5.5.

The practical takeaway is simple, though the benchmark gap by task type is worth knowing before you standardize. If your team has already standardized on one lab’s models for other tools, Claude Code (Anthropic-only) or a Gemini-first Antigravity setup will feel consistent with the rest of your stack, and Gemini specifically pulls ahead on agentic-reasoning and screen-understanding tasks: Agentpedia’s March 2026 testing put Gemini 3.1 Pro at 77.1% on ARC-AGI-2 against roughly 40% for Claude Opus 4.6, and at 72.7% on the UI-focused ScreenSpot-Pro benchmark against roughly 38% for Claude Opus 4.6. If you want the flexibility to A/B test model quality on a hard bug, Antigravity and Cursor both let you switch mid-project, while Claude Code does not give you that option at all.

Gemini 4 Argon Pricing and Antigravity Access

Google announced Gemini 4 Argon on October 13, 2026, with an introductory API price of $2 per million input tokens and $10 per million output tokens, and cached input tokens at 95% off. Google’s own page states that after the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply. That higher price applies to the Argon model intended for public access only. Anyone budgeting API work on Argon should therefore cost it at both rates instead of assuming the introductory price will last.

For Antigravity users, Google’s announcement says Gemini 4 Argon is also available in Antigravity for Google AI Ultra subscribers. The model picker shown on Google’s pages lists Gemini 4 Argon, Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5, and GPT-OSS 120B, so one Antigravity seat can compare Google, Anthropic, and open-weight models on the same task. Google’s pricing page says you can start at $0/month on the Free plan, or upgrade to Google AI Plus, Pro, or Ultra for higher rate limits and optional AI credit overages.

The practical consequence is that model access inside Antigravity now depends on the tier you hold. A solo developer on the Free plan should check which models the picker actually offers before building a workflow around one of them, while a team weighing Argon against Claude Opus 5.5 or Gemini 3.8 Flash can run the same bug through each and compare cost against quality. Because the introductory Argon rate ends on November 30, 2026, any such comparison is most useful when it also prices the post-introductory rate.

Benchmarks: What the Numbers Actually Say

Independent, apples-to-apples benchmarks that test Antigravity, Cursor, and Claude Code as complete products, rather than just the models underneath them, are still thin on the ground in mid-2026, though the underlying model-level numbers have piled up fast. Testing platform Codegen reported in June 2026 that Claude Opus 4.8 scored 78.9% on Terminal-Bench 2.1 against 76.2% for Antigravity’s Gemini 3.5 Flash, and an even wider gap on SWE-bench Verified, 88.6% for Claude Opus 4.8 versus 80.6% for the Gemini 3.1 Pro model running inside Antigravity. That same Codegen report noted the access cost behind those scores barely differs: Antigravity’s CLI rides on Google AI plans starting at $19.99 a month, while Claude Code’s own Pro tier begins at $20 a month, a gap far too small to explain the performance difference on its own, and one that had been much narrower earlier in the year: AugmentCode’s January 2026 numbers had Gemini 3.1 Pro essentially tied with Claude Opus 4.6 Thinking on that same SWE-bench Verified test, 78.8% to 78.2%. What does exist beyond model-level scores is a mix of vendor disclosures and third-party reviewer testing. Treat the numbers below as directional, not as a lab-certified scoreboard.

Metric	Result	Tool / Model	Source type
SWE-bench Verified	High 80s percent range	Anthropic’s latest Claude model family (powers Claude Code)	Independent benchmark trackers, 2026
Gemini coding benchmark claim	~4x faster than prior Gemini generation	Gemini 3.8 Flash (available inside Antigravity)	Google developer blog
Autocomplete acceptance rate	~72%	Cursor	Third-party reviewer comparison
Code retention in final submissions	~88%	GitHub Copilot	Third-party reviewer comparison
Context window (claimed)	200K tokens	Cursor	Vendor documentation
Fortune 100 adoption	~90%	GitHub Copilot	Reported enterprise usage figures

Two caveats matter here, and the benchmark landscape has only gotten noisier since. First, SWE-bench Verified and similar benchmarks test the underlying model, not the surrounding product, so a high score tells you Claude’s reasoning is strong, not that Claude Code’s terminal workflow will feel faster than Cursor’s Composer mode for your specific repo, and reviewers do not always agree on the model-level numbers either: in July 2026, reviewer AIToolBolt measured Claude Sonnet 5 at 80.4% on Terminal-Bench 2.1 against 76.2% for Antigravity’s Gemini 3.5 Flash, a gap running the opposite direction from some other trackers. Second, the acceptance-rate and retention-rate figures for Cursor and Copilot come from third-party comparison sites rather than a controlled academic study, so they are useful for spotting a trend, not for citing as peer-reviewed fact; on the newer SWE-bench Pro benchmark, Kommunicate’s June 2026 testing put Gemini 3.5 Flash at 55.1% against GPT-5.5’s 58.6%, while on SWE-bench Verified the same firm had Claude Sonnet 4.6 slightly ahead of GPT-5.5, 79.6% to 78%. Google’s own claim that Gemini 3.5 Flash runs roughly four times faster than its prior generation, made on its developer blog, is a vendor number and should be read that way too.

Agent Mode and Autonomous Workflows Compared

“Agent mode” means something slightly different in each product, and that difference is often the actual reason a team picks one tool over another.

Antigravity’s agent manager can plan a multi-step task, edit code, open a terminal to run commands, and even control a browser to check that a UI change actually rendered correctly, all without a human clicking through each step. That browser-control piece is genuinely unusual among the three, and it runs fast: OrbilonTech’s June 2026 measurements clocked Antigravity’s Gemini 3.5 Flash at roughly 289 tokens per second, well ahead of the 67 tokens per second the same test recorded for Claude Opus 4.8, though speed and accuracy within Google’s own Antigravity lineup vary by model version, since MorphLLM’s March 2026 testing had the original Antigravity Gemini 3 Pro model scoring 76.2% on SWE-bench Verified versus 80.6% for the newer Gemini 3.1 Pro. Cursor’s Composer mode focuses tightly on multi-file code edits inside the editor itself: describe a refactor, and it touches every file that needs to change, with a review step before anything commits. Claude Code skips the graphical layer entirely. You describe a task in the terminal, it reads the relevant files, writes code, runs your test suite, and reports what it did in plain text, which developers who live in the terminal already find faster than clicking through diffs.

None of these are fire-and-forget yet. Google’s own developer forum has threads describing stability issues and context-memory errors in Antigravity, the kind of rough edges expected from a product still in public preview. Anthropic, for its part, drew criticism reported by Business Insider over confusing 2026 billing changes to Claude Code, separating programmatic API usage from the Max subscription tiers in a way that caught some users off guard. Dev.Classmethod’s July 2026 teardown of Claude Code v2.1.219 found the confusion wasn’t really about the rate itself, since API usage still billed at $5 per million input tokens and $25 per million output tokens, unchanged from the Opus 4.8 pricing it inherited, but about which tasks silently drew from that API meter instead of a user’s Max subscription allowance. Anthropic has since added a fast mode for Opus 5.5, a high-speed configuration with lower-latency responses at a higher cost per token, which is in research preview on Claude Code and priced at $8/$40 per million tokens. Autonomy is improving fast in all three products, helped along by concrete feature drops rather than just model upgrades: Claude Code’s Remote Control and Background Agent capabilities went GA in February 2026, letting tasks keep running unattended, and Computer Use followed on March 23, 2026 for Pro and Max subscribers, per Builder.io, giving the agent a way to operate a full desktop rather than just files and a terminal. Even so, “autonomous” in mid-2026 still means “needs a human checking the output,” not “unsupervised.”

Developer Experience and Learning Curve

Raw capability is only half the decision. How fast a team actually gets productive matters just as much, and the three tools ask for very different amounts of patience up front.

Cursor has the shallowest learning curve of the three for anyone coming from VS Code, since the interface is close enough to the original that muscle memory carries over on day one. Extensions, themes, and keybindings mostly just work. Its downside is the opposite of a learning curve problem: with three years of features stacked on top of each other, new users can feel overwhelmed by how many different ways there are to invoke the AI, from inline edits to chat to Composer mode, before settling on a habit that sticks.

Claude Code asks for more up front. A developer who has never worked comfortably in a terminal is going to fight the tool before they get anything out of it, and there is no fallback graphical mode to lean on when a command-line workflow feels unfamiliar. For engineers who already default to a terminal for git, tests, and deployment scripts, though, the learning curve is closer to zero, since Claude Code just becomes another command they run. Anthropic’s documentation is thorough, but community tutorials and third-party guides are still catching up compared to the multi-year head start Copilot and Cursor have.

Antigravity sits in an odd middle position. Because it launched less than a year ago, in November 2025, its user base and community knowledge base are the thinnest of the three. Stack Overflow threads, YouTube walkthroughs, and internal company playbooks simply have not had time to accumulate the way they have for a tool that has existed since 2023. That said, Google’s own documentation is detailed, and the interface leans on task descriptions in plain language rather than requiring developers to learn a new set of keyboard shortcuts, which flattens the curve for people who are newer to AI-assisted coding generally, even if it is rougher for teams wanting deep customization.

Five Real-World Use Cases

Specs and benchmarks only go so far. Here is how the choice plays out in five situations engineering teams actually face.

• A solo indie developer shipping a side project. Antigravity’s free tier starts at $0, and the ability to try Gemini models on the same bug without paying for several separate subscriptions is a real advantage when every dollar matters.
• An enterprise platform team standardizing tooling for 400 engineers. GitHub Copilot’s existing GitHub integration, admin controls, and $19-to-$39-per-user pricing make it the path of least resistance when procurement and security review are the real bottleneck, not raw coding quality.
• A startup doing heavy multi-file refactors on a large monorepo. Cursor’s Composer mode and 200K-token context window are built exactly for this, letting a small team restructure a codebase without losing track of every file that needs a matching change.
• A backend engineer who lives in the terminal and never opens a GUI editor. Claude Code fits that habit directly instead of forcing a workflow change, and its test-running, file-editing loop matches how many senior engineers already work.
• A team that wants an agent to verify its own UI changes. Antigravity’s browser-control feature lets an agent open the app it just modified and check that a button actually renders where it should, a step every other tool here still leaves to a human.

Migration Guide: Moving Between Antigravity, Cursor, and Claude Code

Switching tools mid-project is more common than vendors like to admit, usually triggered by a pricing change, a model preference, or a new hire who is faster in a different tool. None of these three lock your code into a proprietary format, since the underlying files are just plain source code, but the configuration and agent instructions do not transfer automatically. Here is a practical path.

1. Export or copy any project-specific instruction files first. Cursor stores its rules in a .cursor/rules directory or a legacy .cursorrules file, and both are plain text you can read and adapt.
2. Check for a CLAUDE.md file if you are moving into Claude Code. Anthropic’s CLI reads this file automatically for project-specific context, so porting your Cursor rules into this format preserves most of your existing guardrails.
3. Install the tool you are migrating to. Claude Code installs as a CLI package and authenticates against your existing Claude subscription. Antigravity is available as a downloadable app plus a CLI component after the 2026 update.
4. Re-point any MCP servers you rely on. All three tools support the Model Context Protocol, so a database or API connector configured for one should reconnect in another with the same connection details.
5. Run a small, low-risk task first, a bug fix or a single-file refactor, before trusting the new tool with a large multi-file change. Agent behavior differs enough between these three that the first real task is effectively a calibration run.
6. Keep your old tool’s subscription active for one billing cycle while the team gets comfortable. Cancelling immediately and discovering a gap in agent quality mid-sprint is a common, avoidable mistake.

A quick example of what a Claude Code project context file looks like in practice:

# CLAUDE.md
## Project context
This is a Node.js API using Express and PostgreSQL.
Run tests with: npm test
Lint with: npm run lint
Do not modify files under /legacy without explicit approval.

## Style
Prefer async/await over callbacks.
Match existing error-handling patterns in /src/errors.

That same set of instructions, adapted into a .cursorrules file or an Antigravity project prompt, carries most of the same context across tools, which is why migrating a team’s working conventions takes an afternoon, not a sprint.

Pros and Cons

Google Antigravity

• Pro: Free during public preview, with generous rate limits on a frontier model.
• Pro: Only one of the three that lets you pick between Gemini, Claude, and an open-weight model in the same interface.
• Pro: Browser control lets an agent check its own UI work, not just its code.
• Con: Users report stability issues and context-memory errors on Google’s own developer forum.
• Con: Pricing for heavy usage is bundled into the broader Google AI subscription, making true cost harder to predict.

Cursor

• Pro: Mature Composer mode handles large multi-file refactors well, backed by a claimed 200K-token context window.
• Pro: Backed by a $29.3 billion valuation and heavy investor confidence, which reduces the risk of the product disappearing.
• Pro: Flexible model choice between leading Claude and GPT options.
• Con: The $200 Ultra tier now mainly buys a shared Grok usage pool, with Claude and GPT models billed separately at API list prices.
• Con: Credit-based billing on heavier usage can be harder to predict than a flat per-seat fee.

Claude Code

• Pro: Terminal-native workflow matches how many senior backend engineers already operate.
• Pro: Anthropic doubled usage limits across paid plans on May 6, 2026, easing an early complaint about hitting caps mid-task.
• Pro: Deep integration with a single, consistently updated model family avoids the “which model today” decision fatigue.
• Con: No standalone free tier. You need an active Claude Pro subscription or higher just to start.
• Con: Billing changes in 2026 drew public criticism over confusing separation between subscription and API usage.

Where GitHub Copilot and Windsurf Still Fit In

It would be a mistake to evaluate this category as a three-horse race. GitHub Copilot remains the default answer for large organizations already living inside GitHub, with roughly 90% of Fortune 100 companies using it in some capacity and a Pro plan still priced at just $10 a month, half of what Claude Code charges for its entry tier, though Cursor’s new $8 Plus plan now sits below it. Its June 1, 2026 shift toward usage-based billing on higher tiers briefly paused new Pro and Pro+ signups, but the core value proposition, tight GitHub integration and a very low price, has not changed.

Windsurf’s story is the more dramatic one. In mid-2025, OpenAI was reportedly in talks to acquire Windsurf for roughly $3 billion before the deal collapsed. Google DeepMind then moved in, licensing Windsurf’s technology and hiring CEO Varun Mohan and co-founder Douglas Chen in a deal reported at $2.4 billion on July 11, 2025. Three days later, Cognition AI, the company behind the autonomous coding agent Devin, announced it had signed a binding agreement to acquire what remained of Windsurf, confirmed by TechCrunch the same day. Windsurf still operates as a product under Cognition, and its March 2026 price increase (Pro from $15 to $20 a month, Teams from $30 to $40 a seat) put it on equal footing with Cursor rather than undercutting it, which changes the calculus for teams that picked Windsurf specifically for its lower price.

Security and Enterprise Considerations

Handing an autonomous agent write access to a codebase, a terminal, and in Antigravity’s case a browser, raises the stakes on code review and access control well above what a simple autocomplete tool ever did. Enterprise buyers evaluating any of these three should look at a few specific things: whether the tool offers a business or enterprise tier with admin-level audit logs, whether code sent to the model is used for further training by default, and how mature the company’s incident response has been. GitHub Copilot and Claude Code both offer enterprise tiers with stronger data controls than their free or individual plans. Claude Code’s own enterprise push put a concrete price on that extra scrutiny: Anthropic first shipped a Claude Code Security preview in February 2026 that scans codebases and proposes patches directly, per Zenn.dev, then rolled out automated Security PR reviews for Team and Enterprise users on February 21, 2026, costing roughly $15 to $25 per review according to AIToolsReview UK’s January 2026 coverage of the rollout, giving security teams an actual per-use cost to weigh against manual review time. Antigravity, still in public preview as of mid-2026, has less of a track record here simply because it has existed for a shorter time. Teams in regulated industries should treat any agent with autonomous file-write and command-execution access as a new attack surface requiring its own review, not an extension of an existing code-review process.

Procurement teams should also ask each vendor directly, in writing, three questions before rolling a tool out past a pilot group: what happens to proprietary source code once it is sent to the model, whether that data is retained or used to train future models by default, and whether an admin can revoke a single user’s access without disrupting the rest of the team. GitHub Copilot’s enterprise agreement and Claude Code’s Team and Enterprise tiers both answer these questions in published documentation. Antigravity’s enterprise terms are less mature simply because the product has not existed long enough to build out that layer, which is a reasonable thing to flag in an internal risk assessment rather than a reason to dismiss the tool outright.

Who Should Choose Which Tool

Everything above collapses into a fairly short set of decision points once you strip away the marketing. Budget, existing tooling, and how your team already works together matter more than any single benchmark score. The recommendations below assume you can only commit to one primary tool for now, since running a short pilot with two finalists for a real sprint, rather than a toy project, remains the most reliable way to settle the question for your specific codebase.

• Choose Google Antigravity if budget is the top constraint and you want the flexibility to test Gemini models without paying for several subscriptions.
• Choose Cursor if your team does frequent, large multi-file refactors and values a mature, editor-native agent experience over the lowest possible price.
• Choose Claude Code if your engineers already prefer terminal-first workflows and your organization is standardized on Anthropic’s models elsewhere.
• Choose GitHub Copilot if you need the cheapest reliable entry point and your organization already runs on GitHub for source control and reviews.
• Choose Windsurf if you want a Cursor-like experience from a team now backed by Cognition’s agentic-coding research, and price parity with Cursor does not change your evaluation.
• Avoid all five for now if your codebase is under strict regulatory constraints that prohibit sending source code to third-party model providers, and look at self-hosted alternatives instead.

The Bigger Picture: Why Three Approaches Are Winning at Once

It is tempting to expect one design to eventually win the AI coding category the way one browser engine came to dominate the web. That has not happened, and the reasons are structural rather than temporary. Google, Anysphere, and Anthropic are not actually chasing the same customer. Google can afford to give Antigravity away because the product’s real job is defending Gemini’s position against Anthropic and OpenAI in the broader model war, not generating direct revenue from developer seats. Anysphere has no such luxury. Cursor’s entire business is the editor itself, which is exactly why Anysphere raised The supported figures are the failed $3 billion OpenAI acquisition and the later $2.4 billion Google deal, not a $2.3 billion single round or a pricing comparison to larger-parent-backed tools.[3][10] Anthropic sits in between, using Claude Code partly as a revenue product and partly as a showcase for how well its models handle long, agentic tasks compared to competitors.

That divergence in business motive is exactly why enterprise teams increasingly run more than one of these tools rather than standardizing on a single vendor. A platform team might default to GitHub Copilot for its low per-seat cost and GitHub-native review flow, while a smaller product team inside the same company runs Cursor for a rewrite that needs heavier context, and an individual staff engineer quietly automates deploy scripts with Claude Code because it fits how they already work. None of that requires a company-wide decision, which is part of why adoption of these tools has spread faster than prior generations of developer tooling ever did. The competitive pressure this creates cuts in the user’s favor: every price cut, every doubled usage limit, and every new free tier in 2026 has come from one vendor reacting to what another just shipped.

The Verdict

There is no single winner here, and treating this as a race to one champion misreads the market. For cost-conscious individual developers, Antigravity’s free preview and three-model flexibility currently offer the best value in the category, full stop, as long as you can tolerate the stability rough edges of a public preview product. For teams doing heavy, large-scale refactoring work, Cursor’s Composer mode and its Windsurf’s financing in the provided sources is about $243 million total, with a last known valuation of about $1.25 billion, and the $29.3 billion war chest and Copilot price comparison are not supported.[1][12] For engineers who already live in a terminal, Claude Code removes friction that a GUI-first tool actually adds, even though its lack of a standalone free tier keeps casual users out.

The number that matters most, though, is the one showing up across nearly every tier: $200 a month. Cursor’s Ultra plan, Claude Code’s Max 20x plan, and Windsurf’s top consumer tier have all converged near that price point, though Cursor’s $200 now centers on a shared Grok allowance rather than included Claude and GPT usage. Antigravity undercuts all of them with a free preview, while GitHub Copilot’s $10 Pro plan and Cursor’s $8 Plus plan undercut everyone at the entry level. Pick based on where your team’s actual workflow sits, not on which vendor shouted loudest at its last launch event.

Frequently Asked Questions

Is Google Antigravity actually free to use in 2026?

Yes, for individual developers, Antigravity remains free during its public preview, which started November 18, 2025. Heavier usage draws on the broader Google AI Pro or Ultra subscription tiers, priced between roughly $20 and $200 a month, rather than a separate Antigravity-specific fee.

Can Cursor and Claude Code use the same AI models?

Partially. Cursor lets you choose between premium Claude and GPT models inside its editor. Claude Code only runs Anthropic’s own Claude model family, with no option to switch to a different provider’s model.

Does Claude Code work inside VS Code, or only in the terminal?

Claude Code is terminal-first by design. It runs as a CLI tool alongside whatever editor you already use, rather than living inside one specific IDE window.

Is Google Antigravity built on VS Code?

Multiple secondary reports describe the original Antigravity editor as a fork of VS Code, though Google’s own launch materials describe it more broadly as an agentic development platform rather than confirming that detail directly.

What happened to Windsurf, and is it still an independent company?

No. After OpenAI’s reported $3 billion acquisition bid fell through in mid-2025, Google DeepMind licensed Windsurf’s technology and hired its CEO and a co-founder for roughly $2.4 billion on July 11, 2025. Cognition AI, maker of the autonomous coding agent Devin, then acquired the remaining company on July 14, 2025. Windsurf still operates as a product today, now under Cognition.

Which of these tools has the best free tier for students or hobbyists?

Google Antigravity’s free public preview currently offers the most capability at zero cost, including access to a frontier-class model. GitHub Copilot’s free tier, with 2,000 completions and 50 chat requests a month, is the most established option for developers who want a predictable, long-standing free plan rather than a preview that may change terms later.

Can I switch between these tools without losing my project setup?

Yes, in most cases. Your source code is unaffected since none of these tools use a proprietary file format. Project-specific instructions, stored in files like .cursorrules or CLAUDE.md, need to be manually ported between tools, which typically takes an afternoon rather than a full migration project.

Do any of these tools support the Model Context Protocol (MCP)?

Yes, all three, Antigravity, Cursor, and Claude Code, support MCP, which lets them connect to external tools and data sources like databases or internal APIs using a shared connection standard.

How often do these tools actually change their pricing?

More often than most software categories. Windsurf repriced in March 2026, GitHub Copilot shifted higher tiers to usage-based billing on June 1, 2026, and Anthropic doubled Claude Code’s usage limits on May 6, 2026 without changing the sticker price. Treat any pricing table in this category, including this one, as a snapshot rather than a permanent reference, and check each vendor’s own pricing page before committing a team budget to one tool.