Three of the biggest names in cloud computing now sell their own take on the AI-native code editor, and none of them look alike. Google shipped Antigravity as a free public preview in November 2025. Amazon followed with Kiro, a “spec-driven” agentic IDE built on AWS Bedrock. And Windsurf, the editor Cognition bought in a reported $250 million deal after its earlier flirtation with OpenAI collapsed, spent the first half of 2026 tearing up its own pricing model and rebuilding it around Cascade, its in-house orchestration engine. Anyone shopping for an AI coding IDE in September 2026 is choosing between three very different bets on how much of the coding process an agent should own.
This comparison breaks down pricing tiers, model access, credit systems, and the practical differences that show up once you actually open a terminal and ask one of these tools to ship a pull request. We pulled every number from vendor pricing pages and independent tool trackers current as of September 2026, and we flag anywhere the public data has gaps rather than filling them in with guesses.
What makes this trio worth a dedicated look, rather than folding it into the existing pile of AI-editor comparisons, is who is selling them. Cursor, Windsurf’s pre-Cognition predecessor Codeium, and GitHub Copilot were all built by companies whose entire business is the coding tool itself. Antigravity and Kiro are side bets from Google and Amazon, two companies that would rather you spend more time and more compute inside their existing cloud ecosystems. That changes the incentive structure in ways worth watching: a cloud vendor can afford to give away an IDE for years if it drives Gemini API or Bedrock consumption, in a way a standalone coding-tools startup cannot.
How the Big Cloud Vendors Entered the IDE Wars
Agentic coding tools were, until roughly 2024, the province of venture-backed startups: Cursor, the original Codeium/Windsurf, Replit, and a wave of smaller entrants. That changed fast. Anthropic’s own Claude Code, OpenAI’s Codex CLI, and now Google’s Antigravity and Amazon’s Kiro all represent frontier AI labs and hyperscalers deciding that owning the editor, not just the model API behind it, is worth the investment. Our earlier coverage of Antigravity against Cursor and Claude Code found the same free-vs-paid tension that shows up again here: Google’s preview pricing undercuts everyone, but preview pricing is not the same commitment as a public roadmap.
Amazon’s timing with Kiro follows a similar logic to its historical playbook with AWS services generally: let independent vendors prove out a category, then ship a deeply integrated first-party version that undercuts them on procurement friction even if it doesn’t undercut them on price. We broke down that dynamic in more detail when we compared Kiro directly against Cursor and again in our Kiro vs Claude Code comparison, both of which found Kiro’s spec-driven workflow to be its clearest point of technical differentiation rather than raw model quality, since Kiro leans on the same underlying Claude models available elsewhere.
Windsurf’s path here is the odd one out precisely because it didn’t start as a hyperscaler side project. It has now effectively become one by proxy: Cognition itself has taken venture funding and built its identity around the Devin autonomous agent, and the Windsurf acquisition folded an independent editor into that same orbit. Our prior look at Windsurf against Cursor and GitHub Copilot and Cursor vs Windsurf vs Zed both predate the March 2026 pricing overhaul, which is exactly why this new comparison exists — the numbers in those earlier pieces are no longer current for Windsurf specifically.
What Is Google Antigravity?
Google Antigravity launched in public preview on November 18, 2025, positioned by Google’s own announcement post as “a new era in AI-assisted software” development rather than a simple Cursor clone. The core pitch is an agentic coding environment where the model does more than autocomplete: it navigates a codebase, runs shell commands, and proposes multi-file refactors on its own initiative, with a human reviewing the result rather than typing every line.
What sets antigravity ide apart from a typical vendor-locked assistant is model choice. Gemini 3 Pro is the default agentic backbone, but the same workspace also exposes Anthropic’s Claude Sonnet 4.5 and an open-weight model Google calls GPT-OSS, all inside one interface. That is a deliberate departure from how Google has historically shipped developer tools — bundling a rival’s frontier model into a first-party IDE is not something Microsoft has done with GitHub Copilot or Cursor has done outside its own multi-model picker.
Pricing is the simplest of the three to describe because there is, as of this writing, only one published tier: Individuals: $0 per month, explicitly described on Google’s own pricing page as a way to “experience Antigravity without a subscription plan.” Usage is rate-limited against the underlying Gemini 3 Pro API, and Google has not published team or enterprise pricing for Antigravity separately from standard Gemini API billing. For context, the Gemini 3 Pro API itself (the meter Antigravity draws against once preview rate limits kick in) costs $2 per million input tokens and $12 per million output tokens for prompts under 200,000 tokens, rising to $4 and $18 respectively above that threshold, according to Google’s published Gemini API pricing.
What Is AWS Kiro?
Amazon’s entrant brands itself flatly as a “Spec-Driven Agentic IDE,” and that label is the whole product thesis. Instead of jumping straight from prompt to generated code, kiro ide pushes developers to draft a specification first — what the feature should do, its constraints, its edge cases — and then generates and validates code against that spec. It is the most process-heavy of the three tools, aimed squarely at teams who have been burned by agents that “helpfully” reinterpret a vague prompt into the wrong feature.
Kiro runs on AWS Bedrock, and its model lineup is built around Anthropic’s Claude family rather than a house-built model. The free tier pairs open-weight models with Claude Sonnet 4.5; paying tiers unlock Claude Sonnet 4.6 and Claude Opus 4.8, plus an “Auto” routing option that picks a model per task. That Bedrock foundation matters for enterprise buyers already running workloads inside AWS: procurement, billing, and data-residency conversations can piggyback on an existing AWS relationship instead of opening a new vendor contract.
Kiro’s pricing is the strictest of the three — a flat, credit-metered ladder with five tiers, according to AWS’s published product listing and independent tracker Agentic Index: Free ($0, 50 credits/month), Pro ($20/month, 1,000 credits), Pro+ ($40/month, 2,000 credits), Pro Max ($100/month, 5,000 credits, added in June 2026), and Power ($200/month, 10,000 credits). Every paid tier works out to the same $0.02 per credit, which makes Kiro’s cost curve unusually easy to forecast compared to the other two tools on this list.
What Is Windsurf?
Windsurf has had the most turbulent 18 months of any product in this comparison. It was the editor OpenAI reportedly tried to acquire for roughly $3 billion in mid-2025 before the deal fell apart, and it ended up instead going to Cognition — the company behind the autonomous coding agent Devin — in a deal reported at $250 million. Since that acquisition closed, windsurf ide has absorbed pieces of Cognition’s own stack: Devin Cloud access now ships as a perk on higher tiers, and the two products increasingly share infrastructure.
Technically, Windsurf’s core is Cascade, its agentic orchestration layer that coordinates multi-step tasks like searching a repo, drafting a plan, editing files, and running tests without the developer babysitting each step. On top of Cascade sit two proprietary code models, SWE-1.5 and the newer SWE-1.6, which Windsurf runs at different token throughput depending on plan: roughly 200 tokens per second on the free tier versus 950 tokens per second on Pro, according to a 2026 review that verified the figures against Windsurf’s own product pages.
Windsurf’s pricing has moved twice in the past year, which is worth flagging for anyone comparing screenshots from different months. Before March 19, 2026, the structure was credit-capped: Free (25 credits), Pro at $15/month (500 credits), Teams at $30/user/month, and Enterprise around $60/user/month. On March 19, 2026, Windsurf overhauled that system — what reviewers have taken to calling “Windsurf 2.0” — into a usage-based model: Free ($0, roughly 25 credits/month), Pro ($20/month, with existing subscribers from before the change grandfathered at $15), Max ($200/month), Teams ($40/user/month), and Enterprise at custom pricing. Multiple independent sources, including Codepick’s Windsurf 2.0 deep dive and AICoderscope’s review verified against Windsurf’s site in May 2026, confirm this newer structure.
Antigravity vs Kiro vs Windsurf: Full Specs Comparison
Here is how the three stack up across the technical dimensions that actually change day-to-day usage. Where a vendor has not published a figure, we say so rather than estimate.
| Spec | Google Antigravity | AWS Kiro | Windsurf (Cognition) |
|---|---|---|---|
| Owner / parent | Amazon Web Services | Cognition (acquired 2025-2026, reported $250M) | |
| Public launch | Public preview, Nov 18, 2025 | Live product with tiers as of 2026 | Pricing overhaul “Windsurf 2.0,” March 19, 2026 |
| Core positioning | Multi-model agentic IDE | Spec-driven agentic IDE | Agent orchestration IDE (Cascade) |
| Primary model | Gemini 3 Pro | Claude Sonnet 4.6 / Opus 4.8 (Bedrock) | SWE-1.6 (proprietary) |
| Alternate models available | Claude Sonnet 4.5, GPT-OSS | Open-weight models, Claude Sonnet 4.5 | “All premium models,” Devin integration |
| Compute backbone | Google Gemini API | AWS Bedrock | Cognition’s own infrastructure |
| Free tier | Unlimited months, rate-limited | $0, 50 credits/month | $0, ~25 credits/month |
| Entry paid tier | Not published | $20/mo, 1,000 credits | $20/mo (usage-based, extra billed at API price) |
| Top individual tier | Not published | Power: $200/mo, 10,000 credits | Max: $200/mo |
| Team tier | Not published | Per-user credit tiers apply | Teams: $40/user/mo |
| Enterprise tier | Not published | Not separately published | Custom pricing |
| Token throughput (free vs paid) | Not disclosed | Not disclosed | ~200 tok/s free vs ~950 tok/s Pro (SWE-1.6) |
| Spec-driven workflow | No | Yes (core differentiator) | No (Codemaps for navigation instead) |
| Autonomous agent product tie-in | None disclosed | None disclosed | Devin Cloud access (Max/Teams) |
| GitHub star count / open source | Not open source; no public repo | Not open source; no public repo | Not open source; no public repo |
| Published SWE-bench score | Not published as of Sept 2026 | Not published as of Sept 2026 | Not published as of Sept 2026 |
The row worth staring at is the last one. None of the three companies has put an official SWE-bench Verified or comparable agentic-coding benchmark score on their own pricing or product pages, unlike Anthropic, OpenAI, and several standalone coding-agent vendors who publish those numbers prominently. That is a meaningful omission for buyers trying to compare raw coding capability apples-to-apples, and it means any third-party benchmark claim you see quoted for antigravity ide, kiro ide, or windsurf ide should be treated as an independent lab’s methodology, not a vendor-verified figure.
Pricing Breakdown: Every Tier, Side by Side
Sticker price alone hides how differently these three companies think about monetization. Antigravity is entirely free during its preview window with no announced upgrade path. Kiro is the most rigid: a fixed $0.02 per credit no matter which paid tier you buy, so a team can model its Kiro spend almost like a phone plan. Windsurf abandoned strict credit caps on its paid tiers in the March 2026 overhaul in favor of “unlimited, with overage billed at API price,” which is friendlier for heavy users but harder to budget against in advance.
| Tool | Tier | Price | Monthly Allowance | Models Included |
|---|---|---|---|---|
| Antigravity | Individuals | $0/mo | Rate-limited (no published cap) | Gemini 3 Pro, Claude Sonnet 4.5, GPT-OSS |
| Kiro | Free | $0/mo | 50 credits | Open-weight models, Claude Sonnet 4.5 |
| Kiro | Pro | $20/mo | 1,000 credits | Auto, Claude Sonnet 4.6, Claude Opus 4.8 |
| Kiro | Pro+ | $40/mo | 2,000 credits | Same as Pro |
| Kiro | Pro Max | $100/mo | 5,000 credits | Same as Pro |
| Kiro | Power | $200/mo | 10,000 credits | Same as Pro |
| Windsurf | Free | $0/mo | ~25 credits | SWE-1.6 (200 tok/s), Tab completions |
| Windsurf | Pro | $20/mo (new); $15/mo grandfathered | Unlimited, overage at API price | SWE-1.5, all premium models |
| Windsurf | Max | $200/mo | Unlimited, overage at API price | Devin Cloud access |
| Windsurf | Teams | $40/user/mo | Unlimited, overage at API price | SSO, RBAC, admin dashboard |
| Windsurf | Enterprise | Custom | Custom | Hybrid deployment, all Teams features |
Do the math at the top end and the framing behind our headline number gets clearer: Kiro’s Power tier and Windsurf’s Max tier both land at $200 a month for a single power user, while Antigravity’s entire published pricing page still reads $0. That is not proof Antigravity is “worth” more or less than the other two — it is a preview-stage product without a monetization model yet, and Google has not said what happens when the free preview ends. But for a team doing cost planning today, it is the single most actionable number in this whole comparison.
Model Access and Underlying AI Benchmarks
Because none of the three vendors publishes an end-to-end coding benchmark for their IDE product itself, the closest thing to an apples-to-apples comparison is looking at what models power each tool and what is publicly known about those models individually. This is an important distinction: a benchmark score for “Gemini 3 Pro” measures the underlying model, not how well Antigravity’s agent loop, tool-calling, and context management wrap around it — two products built on the same model can still feel very different to use.
Antigravity’s default model, Gemini 3 Pro, is billed through the standard Gemini API tiering — cheaper below a 200,000-token prompt, more expensive above it — which implies a context window comfortably in the hundreds of thousands of tokens, though Google has not published an exact Antigravity-specific ceiling. Kiro’s top-tier models, Claude Sonnet 4.6 and Claude Opus 4.8, are accessed through AWS Bedrock rather than Anthropic’s API directly, which means Kiro users inherit Bedrock’s regional availability and data-handling commitments rather than Anthropic’s own. Windsurf is the outlier: SWE-1.5 and SWE-1.6 are proprietary models trained by Cognition specifically for coding tasks, not general-purpose frontier models repurposed for code, and Windsurf backs them up with “all premium models” access for developers who want a second opinion from a non-Cognition model.
The one hard performance number in the public record is Windsurf’s token throughput split: roughly 200 tokens per second on the free SWE-1.6 tier versus roughly 950 tokens per second on Pro. That 4.75x throughput gap is the most concrete, vendor-adjacent performance figure available across all three products, and it is worth noting that neither Antigravity nor Kiro discloses an equivalent number for their own agent loops.
How the Agent Loop Actually Works in Each Tool
The phrase “agentic IDE” gets used loosely across the industry, so it’s worth being specific about what each of these three products actually does when you hand it a task, based on how the vendors themselves describe the workflow.
In Antigravity, a developer describes a goal in natural language and the agent — running on Gemini 3 Pro by default — plans a sequence of actions, reads relevant files, and can execute shell commands directly rather than only suggesting a diff for approval. Because the same session can switch models, a developer can ask Claude Sonnet 4.5 to review a plan that Gemini 3 Pro generated, effectively getting a second opinion from a different lab’s model without leaving the editor. That cross-model review pattern is the single most distinctive workflow feature Antigravity offers that neither Kiro nor Windsurf currently replicates in the same way.
Kiro’s loop is intentionally slower at the front end. Before any code gets generated, the developer (often with the agent’s help) writes a spec: what the feature does, what it explicitly does not do, and what constraints apply. Only after that spec exists does Kiro generate code and then validate the output against the spec it just wrote. This is closer to how a senior engineer might onboard a junior teammate — write the ticket properly before anyone touches a keyboard — and it is the reason Kiro tends to get recommended for teams with compliance or correctness requirements rather than for rapid prototyping.
Windsurf’s Cascade engine sits somewhere between the two. It orchestrates multi-step tasks — searching the repo, building a plan, editing across files, running tests — with the SWE-1.5 or SWE-1.6 model doing the actual code generation, and Codemaps giving the agent (and the human reviewing its output) a navigable map of how a change touches the broader codebase. At the Max and Teams tiers, a developer can also hand a task off to Devin Cloud for longer-running, less-supervised work, which is a meaningfully different mode from Cascade’s more interactive loop — closer to filing a ticket for an async teammate than pairing with an agent in real time.
Cost Modeling: Three Team Budgets Compared
Published tier pricing only tells part of the story — what matters for a budget conversation is what a realistic team actually pays once seats and usage stack up. Using each vendor’s own published tier pricing, here is how the numbers land for three common team sizes, assuming each team picks the tier that comfortably covers a typical workload (Pro-equivalent for solo/small teams, Teams-equivalent tiers where available for larger groups).
| Team Size | Antigravity | Kiro | Windsurf |
|---|---|---|---|
| 1 developer (solo) | $0/mo (free preview) | $20/mo (Pro, 1,000 credits) | $20/mo (Pro, usage-based) |
| 5-person team | $0/mo (free preview, no team billing yet) | $100/mo (5 x Pro at $20 each) | $200/mo (5 x Teams at $40 each) |
| 20-person team | Not available (no published team tier) | $400/mo (20 x Pro at $20 each) | $800/mo (20 x Teams at $40 each) |
| Heavy individual power user | Capped by preview rate limits | $200/mo (Power, 10,000 credits) | $200/mo (Max, unlimited + overage) |
Two things stand out from that table. First, Antigravity’s lack of a published team tier is a real gap for any organization past the solo-developer stage — there is currently no way to centrally bill, manage seats, or apply admin controls across a team, which pushes any company larger than a few developers toward Kiro or Windsurf almost by default, regardless of model preference. Second, at the 20-person mark, the gap between Kiro ($400/month) and Windsurf ($800/month) is entirely explained by Windsurf’s higher per-seat Teams price ($40 vs Kiro’s $20 Pro tier), not by usage — a detail that matters more than headline “Free tier” comparisons once an organization is actually paying for seats.
Real-World Use Cases: Where Each Tool Fits
None of these three products has published named customer case studies as of September 2026 — they are simply too new, and none of the vendors has released the kind of “Company X shipped Y using our tool” material that older products like GitHub Copilot have accumulated over several years. What we can lay out instead are the concrete workflow scenarios each tool is architecturally built for, based on how the vendors themselves describe their products.
- A startup already deployed on AWS Bedrock evaluating an agentic IDE for its backend team gets the cleanest procurement path with Kiro, since billing, IAM, and compliance controls already live inside the same AWS account rather than a separate vendor.
- A team that has been burned by an agent “guessing” the wrong feature from an underspecified prompt is the textbook case for Kiro’s spec-driven workflow, which forces a written spec before code generation begins.
- A solo developer or small team wanting to compare model output without paying for three separate subscriptions benefits most from Antigravity’s free preview, which bundles Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS in one workspace at no cost.
- An engineering org already running Devin for autonomous, long-running tasks gets the most direct integration story from Windsurf’s Max and Teams tiers, which include Devin Cloud access as part of the same Cognition product family.
- A budget-conscious freelancer who needs predictable monthly costs is better served by Kiro’s flat $0.02-per-credit pricing than by Windsurf’s usage-based overage model, which can vary month to month depending on how much a project leans on the agent.
- A team standardized on Anthropic’s Claude models for other parts of its stack has two viable on-ramps here: Kiro (via Bedrock) or Antigravity (via direct Claude Sonnet 4.5 access), depending on whether the team already has AWS or Google Cloud billing in place.
Pros and Cons of Google Antigravity
Antigravity’s biggest advantage is also its biggest question mark: it is free. For evaluation purposes, that makes it the lowest-friction of the three tools to try, and the multi-model picker means a developer can compare Gemini 3 Pro against Claude Sonnet 4.5 on the exact same task without switching apps. The downside is that “public preview” is not a permanent commitment. Google has not said what pricing looks like once the preview period ends, which makes Antigravity a riskier long-term bet for a team building workflows around it today. The product also has the thinnest public documentation of the three when it comes to enterprise features — no published SSO, RBAC, or team-billing story yet, which rules it out for larger organizations that need those controls now rather than eventually.
Pros and Cons of AWS Kiro
Kiro’s spec-driven approach is a genuine differentiator, not just marketing language — forcing a written spec before code generation is a workflow discipline that catches a class of “agent went off and built the wrong thing” failures that pure chat-to-code tools do not. The flat $0.02-per-credit pricing across every paid tier also makes Kiro the easiest of the three to budget precisely. The tradeoffs: the spec-first workflow adds upfront overhead that will feel like friction on small, throwaway tasks, and Kiro’s tightest coupling is to AWS Bedrock, which is a selling point for AWS-native shops and a mild lock-in concern for teams that want to stay cloud-agnostic. Kiro also has not published a context window figure or an independent benchmark score, so buyers have to lean on Claude’s known model-level capabilities rather than Kiro-specific numbers.
Pros and Cons of Windsurf
Windsurf is the most mature product of the three in terms of published enterprise features — SSO, RBAC, centralized billing, and an admin dashboard are all live today on Teams and Enterprise tiers, not just promised. The Devin Cloud integration is also unique in this comparison: no other tool here connects an IDE directly to an autonomous long-running coding agent from the same company. The downside is pricing whiplash. Windsurf changed its entire pricing model in March 2026, and teams that priced out the product before that date were working from numbers that no longer apply — a reminder to always check the vendor’s live pricing page rather than trusting a cached screenshot, including the tables in this article. The acquisition by Cognition is also relatively recent, and buyers evaluating long-term platform stability should factor in that Windsurf has changed ownership once already in the past two years.
Migration Guide: Switching Between the Three
Moving an existing workflow from one of these tools to another is less about data portability — none of the three locks your source code into a proprietary format — and more about rebuilding muscle memory around how each tool wants to be prompted. Here is a practical path for a team migrating in either direction.
- Export or document any custom rules, prompt templates, or “memories” you have built up in your current tool, since none of the three currently offers a direct one-click import from a competitor.
- If moving toward Kiro, start by drafting specs for your two or three most complex in-flight features before touching the code generation step — this is the habit the whole tool is built around, and skipping it defeats the point.
- If moving toward Antigravity, take advantage of the free multi-model access during your evaluation window to A/B test Gemini 3 Pro against Claude Sonnet 4.5 on the same ticket before standardizing on one.
- If moving toward Windsurf, confirm which pricing era your quote reflects — pre- or post-March 19, 2026 — since older $15 Pro pricing is grandfathered and not available to new subscribers.
- Re-point your CI/CD and repo access: all three tools need fresh authentication against your Git provider, and Kiro specifically needs IAM permissions configured if you are integrating it with existing AWS infrastructure.
- Rebuild your team’s linting and formatting expectations into whichever tool’s rules or spec system you land on, since agent-generated code without house style guardrails tends to drift fast across a team.
- Pilot with a single team or a single repo for two to four weeks before a full rollout, and track how many agent-generated pull requests need substantial human rework versus how many merge close to as-generated.
- Decide on a credit or usage budget per seat before rollout — Kiro’s flat per-credit pricing makes this easy to model, while Windsurf’s usage-based overage requires watching actual spend for a full billing cycle first.
- If your organization needs SSO, RBAC, or audit logging today rather than “eventually,” confirm those features are live on your target tier — as of September 2026 that rules out Antigravity for compliance-sensitive teams.
- Keep a fallback path open. Given how fast pricing and feature sets have moved on all three products in under a year, avoid deep, hard-to-reverse integrations until the market settles.
Security, Compliance, and Enterprise Readiness
Enterprise buyers will find the most daylight between these three products here. Windsurf’s Enterprise tier is explicitly built around SSO, RBAC, and hybrid deployment options, and earlier 2026 vendor materials referenced SOC 2, HIPAA, and FedRAMP compliance targets for that tier, alongside audit controls carried over from the pre-acquisition Codeium product. Kiro inherits its compliance posture largely from AWS Bedrock itself, which already carries a broad set of AWS compliance certifications that Bedrock-based products can lean on, though Amazon has not published a Kiro-specific compliance page separate from standard AWS documentation. Antigravity, still in public preview, has the thinnest enterprise story of the three: no published SSO, RBAC, or dedicated compliance documentation as of September 2026, which is consistent with a product still finding its monetization and governance model rather than one actively courting large enterprise contracts.
Which One Should You Choose?
The honest answer depends less on which tool is “best” in the abstract and more on which cloud ecosystem, budget model, and workflow discipline your team already has. Choose Antigravity if you want zero-cost access to three frontier-class models in one workspace and you are comfortable with the uncertainty of a preview product that could change its pricing at any time. Choose kiro ide if your infrastructure already lives on AWS, your team has been burned by agents misreading vague requirements, and you want pricing you can forecast to the cent. Choose windsurf ide if you need enterprise-grade access controls today, you want the option of Devin-style autonomous agents down the line, and you are comfortable checking pricing pages carefully given how much the product has changed ownership and cost structure over the past year.
The Verdict
None of these three tools has a published benchmark score to settle an argument about which one writes better code, and anyone telling you otherwise with high confidence in September 2026 is extrapolating past what these vendors have actually released. What the public data does settle clearly is the economics: Antigravity is free with no announced ceiling, Kiro is the most predictable at a flat $0.02 per credit across every paid tier, and Windsurf is the most enterprise-ready but has the most volatile pricing history of the three, having overhauled its entire model in March 2026 alone. For most individual developers and small teams evaluating right now, the rational move is to run all three during Antigravity’s free preview window, since it costs nothing to compare Gemini 3 Pro, Claude Sonnet 4.5, and Kiro’s or Windsurf’s free tiers side by side on the same real ticket before committing a team budget to any of them.
What Could Change This Comparison in the Next Few Months
Every number in this article is a snapshot, and the underlying products have already proven they move fast. The most likely disruption is Antigravity graduating out of public preview — when that happens, whatever paid tier Google introduces will instantly become the most important new data point in this comparison, since it is the only one of the three without any pricing history to extrapolate from. Watch for whether Google prices it against Kiro’s credit model, against Windsurf’s usage-based model, or invents a third structure entirely.
Kiro is the most likely of the three to add a new tier rather than restructure existing ones, based on its track record: Pro Max was added in June 2026 without touching the pricing or allowances of the four tiers that already existed. That suggests Amazon views Kiro’s five-tier ladder as close to its final shape, with room to add rungs at the top rather than rebuild the middle.
Windsurf is the wild card. Having already overhauled its pricing once in the past year and changed ownership once in the past two, betting on stability from Windsurf specifically is the riskiest assumption in this entire comparison. Teams building long-term budget forecasts around windsurf ide should plan to re-verify pricing every quarter rather than treating any single snapshot, including this one, as durable.
Frequently Asked Questions
Is Google Antigravity free to use?
Yes. As of September 2026, Antigravity’s only published pricing tier is “Individuals: $0 per month,” available during its public preview with rate-limited access to Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS. Google has not announced when the preview ends or what a paid tier would cost.
What AI models power AWS Kiro?
Kiro runs on AWS Bedrock and is built primarily around Anthropic’s Claude models. The free tier includes open-weight models plus Claude Sonnet 4.5, while paid tiers from Pro upward unlock Claude Sonnet 4.6 and Claude Opus 4.8.
Who owns Windsurf now?
Windsurf is owned by Cognition, the company behind the autonomous coding agent Devin, in a deal reported at $250 million. This came after an earlier, reportedly larger acquisition attempt by OpenAI fell through in 2025.
What is the most expensive individual plan across all three tools?
Kiro’s Power tier and Windsurf’s Max tier are both priced at $200 per month for a single user. Antigravity has no published paid individual tier as of September 2026.
Does any of these three tools publish a SWE-bench score?
No. As of September 2026, none of Google, AWS, or Cognition/Windsurf has published an official SWE-bench Verified or comparable agentic-coding benchmark score for Antigravity, Kiro, or Windsurf specifically. Any benchmark figures circulating for these products come from third-party evaluations rather than the vendors themselves.
What is Kiro’s “spec-driven” approach?
Kiro requires developers to draft a written specification describing what a feature should do, its constraints, and edge cases before the agent generates code. The code is then generated and validated against that spec, rather than generated directly from a loose prompt.
Can I use more than one model inside the same IDE?
Antigravity is the clearest example — it bundles Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS in one workspace. Windsurf also advertises access to “all premium models” alongside its proprietary SWE-1.5 and SWE-1.6 models. Kiro is more Claude-centric, though its free tier does include open-weight model access alongside Claude Sonnet 4.5.
Which tool has enterprise features like SSO and RBAC available today?
Windsurf currently has the most complete published enterprise feature set among the three, with SSO, RBAC, and centralized billing live on its Teams and Enterprise tiers. Kiro’s enterprise posture leans on AWS Bedrock’s existing compliance certifications. Antigravity has not published enterprise-specific access controls as of September 2026.