Join our Newsletter — 33% off our NHI Course
Home› Guides› AI IDEs Compared: Plans, Models, Platforms and Admin…
Guide AI Platforms

AI IDEs Compared: Plans, Models, Platforms and Admin Controls

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 15 November 2026 17 min read
On this page

AI IDEs now bundle an editor, an agent, a model catalogue and a billing scheme, and each vendor describes them with different words and different units. This page lines up five AI IDEs with the same fields for each: Cursor, Devin Desktop (the IDE formerly called Windsurf), Google Antigravity, Kiro and Zed. It covers plans and prices, models, agent modes and approval settings, supported platforms, enterprise and admin controls, and data handling, all taken from each vendor's own pricing pages and documentation. It is written for developers, engineering managers and the people who approve tools for a team. Tools are listed alphabetically; order is never a ranking, and the page gives no verdicts.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. Plans, prices and model lists change often, so confirm against each provider's current pages before buying.

Key takeaways

  • Five IDEs, five usage units. Cursor states usage as multiples of its Plus plan, Devin Desktop uses quotas that refresh daily and weekly, Antigravity uses weekly rate limits and an AI credit pool, Kiro uses credits with per-model multipliers, and Zed bills hosted tokens at API list price plus 10% beyond an included amount.
  • Entry paid prices run from $8 to $20 a month. Cursor Plus is $8, Zed Pro is $10, and Devin Pro and Kiro Pro are $20; Antigravity paid individual access comes through Google AI Pro and Ultra subscriptions, whose prices are not shown on the Antigravity pricing page.
  • Every tool offers several model providers. All five list Anthropic and OpenAI or Google models alongside, in some cases, the vendor's own models (Cursor Composer and Grok, Cognition SWE models, Gemini) and open-weight models.
  • Team controls sit on team or enterprise tiers. Model allowlists, MCP restrictions, SSO and audit settings appear on Teams, Business or Enterprise tiers in every tool's documentation.
  • Autonomy is a setting, not a fixed property. Each IDE lets users or admins choose how far the agent runs terminal commands and edits without asking, and those defaults are where most team policy decisions sit.

Which AI IDEs are on this page?

The inclusion rule is objective: a desktop editor whose vendor documentation describes a built-in coding agent that can edit files and run commands, with plans confirmed on an official page on 2026-11-15. Editor extensions that add AI to an existing IDE (such as GitHub Copilot) are covered in Cursor vs GitHub Copilot and the AI Coding Tools Hub; terminal agents are in AI Coding CLI Agents Compared; browser app builders are in Vibe Coding Tools Compared.

  • Cursor: an AI code editor with Tab completion, an agent, cloud agents and a command line agent. Latest desktop version on the downloads page: 3.30.
  • Devin Desktop: Cognition's IDE. Its FAQ states "Devin Desktop is the new name for Windsurf"; the rename arrived as an over-the-air update on 2 June 2026, with the same features.
  • Google Antigravity: described in Google's getting-started guide as a suite of agent tools across a desktop app, a standalone IDE, a CLI (agy), IDE extensions and a Python SDK.
  • Kiro: an AWS agentic IDE and CLI built around spec-driven development, and the migration target named for Amazon Q Developer IDE users.
  • Zed: an editor from Zed Industries, open source under GPL-3.0-or-later with Apache-2.0 components where marked, with the native Zed Agent plus external agents over the Agent Client Protocol (ACP).

For single-tool depth, see What Is Cursor? Features, Pricing, Models and History, Devin Desktop (formerly Windsurf), What Is Google Antigravity? Agent-First IDE Explained and What Is Kiro? AWS Spec-Driven IDE, Pricing and Q Developer Migration.

AI IDEs at a glance

As of 2026-11-15, from each vendor's own pages.

FieldCursorDevin DesktopGoogle AntigravityKiroZed
VendorCursorCognitionGoogleAWSZed Industries
SurfacesDesktop editor, cloud agents, CLI, Agent SDKDesktop IDE, Devin Local agent, Devin CloudDesktop app, standalone IDE, CLI, extensions for VS Code, Visual Studio and Eclipse, Python SDKIDE and CLIDesktop editor with Zed Agent, external ACP agents and Terminal Threads
LicenceProprietaryProprietaryProprietary, Google Cloud Terms for businessProprietaryGPL-3.0-or-later, Apache-2.0 components
Free tierFree, limited agent and Chat requestsFree, light quota, unlimited TabIndividuals, $0, basic weekly limits, unlimited TabFree, 50 credits a monthPersonal, $0, bring your own keys or external agents
Rules and project instructionsRules, skills, subagents, hooksAGENTS.md, .devin/rules/, imports .cursor/rulesProjects set repository boundaries, file scopes and permissionsSpecs, steering and hooks (documented sections)Profiles, skills, instructions
MCPYes, admin access controls on EnterpriseYes, admin allowlist and registryYes, MCP tools under permission rulesYes, admin registry controlYes, for Zed Agent

How much do AI IDEs cost?

List prices in US dollars per month, excluding tax, as shown on each pricing page on 2026-11-15. Annual pricing is shown where the vendor shows it; none of the five pages lists annual discounts for these tiers.

ToolIndividual paid tiersTeam tierEnterpriseBeyond the allowance
CursorPlus $8 (1x usage, Grok models only); Premium $30 (4x); Super $100 (15x); Ultra $200 (50x)Teams from $30 per userCustom, pooled usageAllowance refreshes weekly; top-ups; third-party models billed separately at API list price
Devin DesktopPro $20; Max $200Teams $80 a month plus $40 per full dev seat, up to 100 usersCustomExtra usage at API pricing; quotas refresh daily and weekly
Google AntigravityGoogle AI Pro and Google AI Ultra (higher limits and an AI credit pool; price not on the Antigravity page)Business through Google Cloud: Standard and Plus from $30 per seat; Pay-as-you-go with $0 seat feeContact sales; business plans open to eligible Gemini Enterprise customersConsumption pricing on Pay-as-you-go and API; spend caps
KiroPro $20 (1,000 credits); Pro+ $40 (2,000); Pro Max $100 (5,000); Power $200 (10,000)Teams and Enterprise $20 per seat, 500 pooled credits per seatSame seat price, central billing and SSOAdd-on credits $0.04 each; AWS GovCloud (US) about 20% higher
ZedPro $10 (unlimited edit predictions, $5 of tokens)Business $30 per seatOrder forms from 25 seatsAPI list price plus 10%; spending cap available

Trials: Kiro gives the rest of the first month free plus 500 trial credits on the first upgrade to Pro (card required). Zed Pro has a 14-day trial with $5 of usage and no card. Cursor and Devin start with a Free tier; Antigravity lists a trial route for Gemini 4 Argon on eligible premium devices. For cost per developer across a wider tool set, see the AI Coding Tools Pricing Tracker.

Which models does each AI IDE offer?

Model lists as published on each vendor's models page on 2026-11-15. Lists are long and change monthly; the table names providers and representative models rather than every variant.

ToolVendor or in-house modelsThird-party modelsFree plan modelsRouting
CursorComposer 2.5; Grok 4.5, 4.6, 4.7 (listed under Cursor's own models)Anthropic Claude Fable 5.2, Opus 5.5, Sonnet 5.5, Haiku 5.5; Google Gemini 3.1 Pro, 3.8 Flash, 4 Argon, 4 Carbon, 4 Hydrogen; OpenAI GPT-5.6 Sol, Terra, Luna; Meta Muse WonderLimited usage; third-party models on Premium and aboveAuto: Cost mode on all plans; Balance and Intelligence on Teams and Enterprise with Cursor Router
Devin DesktopSWE-2, SWE-2 Flash, SWE-1.7, SWE-1.6, SWE-1-mini (Tab), swe-grep, swe-checkAnthropic Claude Fable 5.2; OpenAI GPT-5.6 Sol; Google Gemini 3.7 Flash, 3.8 Flash, 4 ArgonSWE-2 Flash at Medium effortAdmin default model and allowlist
Google AntigravityGemini 4 Carbon, 4 Argon, 4 Hydrogen, 3.8 Flash; Orchestrator pairingsAnthropic Claude Sonnet 5.5 and Opus 5.5 (thinking); GPT-OSS 120BGemini 4 Hydrogen, Gemini 3.8 Flash and others marked for FreeOrchestrator (Argon or Carbon) on Pro and Ultra
KiroAuto router (1.0x)Anthropic Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and earlier; OpenAI GPT-5.6 Sol, Terra, Luna; DeepSeek 3.2; MiniMax M2.5; GLM-5; Qwen3 Coder Next; Grok 4.7; Kimi K3Auto, Claude Sonnet 4.5 and 4.0, and the open-weight modelsAuto as default; per-model credit multipliers from 0.05x to 6x
ZedZeta for edit predictionZed-hosted models, or your own keys for providers including Amazon Bedrock, Anthropic, DeepSeek, Google AI, Mistral, OpenAI, OpenRouter and Vercel; local models through Ollama and LM StudioPersonal plan uses your own keys or external agentsUser or admin model choice

Retirements: Antigravity lists Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS 120B for removal on 18 November 2026. Kiro's admin docs state that Claude Fable 5.1 and Kimi K3 are unavailable until an admin turns on model access management. For the models themselves, see Coding LLMs Compared.

Agent modes and approval settings

Every IDE here lets the agent edit files and run terminal commands. What differs is the name and the default of the setting that decides when it asks first. The values below are the setting names in each vendor's documentation.

ToolAutonomy settingDocumented defaultIsolation option
CursorAuto-run, browser and network controls (Enterprise, under LLM Safety and Controls)Admin-configured on EnterpriseCloud agents, including on self-hosted machines
Devin DesktopMaximum autonomy level: no auto-approvals, allowlist only, smart, or bypassDevin Local Agent off on Enterprise; web search off for enterprise teamsDelegation to Devin Cloud
Google AntigravityTerminal command auto execution (Request Review or Always Proceed); CLI toolPermission (request-review, strict, always-proceed, proceed-in-sandbox)request-review; Artifact Review asks-for-review; non-workspace access offSandbox Mode (off by default; Seatbelt on macOS, kernel namespaces on Linux); worktree mode
KiroAutopilot or Supervised mode; Trusted Commands listAutopilot mode; web tools on; MCP open; CLI API keys off; cloud sessions off for IAM Identity Center orgsCloud sessions
ZedTool permissions per agent profilePrompts and code context not kept by Zed by defaultAgent sandboxing; external agents run in their own process

Precedence rules matter: Antigravity documents that deny outranks ask and ask outranks allow, in every mode. Devin Desktop documents that once one MCP server is on the allowlist, every unlisted server is blocked. Check equivalent rules in the other tools before writing a team policy. For the threats these settings address, see the AI IDE and Coding Agent Security Guide and Coding Agent Config File Attacks.

Supported operating systems and IDEs

ToolmacOSWindowsLinuxOther surfaces
CursorApple Silicon, Intel, Universalx64 and ARM64, system and user installersx64 and ARM64; .deb, RPM, AppImageCLI install script; JetBrains and Xcode integrations listed in docs
Devin DesktopOS X Yosemite or laterWindows 10 or laterTarball (glibc 2.28+), deb (Ubuntu 20.04+, Debian 10+), rpm (Fedora 36+, CentOS 8+, RHEL 8+)Linux tarball build does not self-update
Google AntigravityApp: macOS 13 or later; IDE: macOS 12 or laterWindows 10 (64-bit) or 11, x64 or ARM64glibc 2.28+ and glibcxx 3.4.25+; preinstalled on Googlebook OSVS Code 1.90+, Visual Studio 2026 (17.8+), Eclipse 2026-06+ (public preview); enterprise sign-in in preview for JetBrains, Zed and Xcode
KiroApple Silicon and Intel (DMG or PKG)ListedListedKiro CLI on macOS, Windows and Linux
ZedListedListedListedPackage managers; web listed as not yet available

Enterprise and admin controls

Platforms are listed alphabetically, each with the controls its own documentation names. Plan and edition limits are in one Availability note per tool.

Cursor

Enterprise settings cover SAML and OIDC SSO, SCIM provisioning, audit logs of authentication, user management and admin actions, model access restrictions, MDM policies that enforce allowed team IDs and extensions, network configuration with proxy and IP allowlisting, and org-wide enforcement of Privacy Mode. The Enterprise tier also lists repository, model and MCP access controls, group-level spend limits and service accounts. Availability: SSO on Teams and Enterprise; SCIM, audit logs and model restrictions on Enterprise.

Devin Desktop

Team settings, set in the Devin app or the Windsurf dashboard, include allowlisted models (by model or by provider), a default model, maximum autonomy level, Allow MCP, an MCP allowlist, MCP registry enforcement, web search, data retention toggles for models that need up to 30 days of stored data, codemap sharing and the Devin Local Agent switch. Device policies through MDM cover AllowedExtensions, ExtensionGalleryServiceUrl, UpdateMode, TelemetryLevel and EnableFeedback. Availability: SAML and OIDC SSO on Teams; SCIM, role-based access control and VPC deployment on Enterprise.

Google Antigravity

Business access runs through Google Cloud and Gemini Enterprise, with centralised admin controls on eligible subscription plans, spend caps across API and subscription plans and coverage under Google Cloud Terms of Service. Business accounts sign in with a Google Cloud business account, licence and project; telemetry in the CLI is always off for business accounts, and business accounts use the earlier permission settings rather than the Permission Preset. Availability: eligible Gemini Enterprise customers; enterprise sign-in generally available in VS Code and Visual Studio.

Kiro

Governance controls cover model access management with an approved model list, MCP (disable entirely or allow only registry servers), Kiro CLI API key generation, web tools, sign-in method restrictions through managed-settings.json (Kiro IDE 1.2+ and Kiro CLI 2.25.0+) and cloud sessions. Admins can set usage limits per user, group or profile, with alerts at 80% and 100%. Availability: Teams and Enterprise at $20 per seat, with IAM Identity Center for organisations.

Zed

Business admin controls let owners restrict Zed-hosted models, turn off edit predictions organisation-wide, lock data sharing settings and see unified spend, with role-based access control. Availability: Business at $30 per seat; as of November 2026 the pricing page lists SSO, SAML and SCIM as planned for Business.

For a wider control-by-control view across more tools, see AI Coding Tools Enterprise Controls.

Data handling and model training

What each vendor's own data page states, as of 2026-11-15. Defaults differ by plan, so check the plan you buy.

ToolTraining useRetention and providersOrg enforcement
CursorWith Privacy Mode on, customer data is not used for training by Cursor; with it off, Cursor may use data to improve and train modelsZero data retention agreements with all providers under Privacy Mode, with abuse-detection exceptions; file contents cached temporarily with request-scoped keys; your own API key requests still pass through Cursor's backendPrivacy Mode enforced org-wide on Teams and Enterprise
Devin DesktopData retention toggle unlocks models that store data for up to 30 days, not used for trainingPer-provider retention toggles for Anthropic and OpenAI modelsTeam settings and TelemetryLevel device policy
Google AntigravityEnable Telemetry collects interactions to evaluate and improve Antigravity and supporting modelsBusiness accounts covered by Google Cloud Terms of ServiceCLI telemetry always off for business accounts
KiroFree and individual: content used for service improvement unless the user opts out; Enterprise: opted out automaticallyUS East (N. Virginia) for Free and individual; profile Region for Enterprise; Free inputs kept up to 60 days for abuse detection; TLS 1.2+, AWS KMS, customer managed keys on EnterpriseAdmin-set telemetry on Enterprise
ZedEdit prediction training data only on opt-in, for open-source projects and non-excluded filesZed does not keep prompts or code context by default; zero data retention required of hosted providers, with a stated exception for Anthropic Covered Models (at least 30 days)Business admins lock data sharing settings

Anthropic Covered Models: Zed's privacy page and Kiro's data page both state that certain Claude Fable models carry provider-side retention of up to or at least 30 days, which applies wherever those models are served. Check the same point for any model you allow in any tool. A per-tool tracker is at AI Coding Tools Data Retention and Training Tracker.

How to choose an AI IDE: factual criteria

  1. Start from the identity system you already run. If you need SCIM today, list the tiers whose documentation names it (Cursor Enterprise, Devin Enterprise) and check the planned or Google Cloud routes for the others. Kiro uses IAM Identity Center; Antigravity business access uses Google Cloud accounts.
  2. Match the model list to your approved providers. If policy restricts providers or open-weight models, check whether the tool offers an admin model allowlist (all five document a form of model restriction on team tiers) and whether restricted models are off by default.
  3. Model the usage unit against a pilot task. Run one identical task in each shortlisted IDE and record the usage consumed: plan multiples, quotas, rate limits, credits or tokens.
  4. Check the autonomy defaults. Record the default for terminal execution, web access and MCP in each tool, and whether an admin can cap it centrally.
  5. Check platform fit. Linux packaging, ARM64 Windows and JetBrains or Visual Studio support vary by tool (see the platform table).
  6. Read the data page for the exact tier. Training defaults differ between free, individual and enterprise tiers in several tools.
  7. Consider source availability where it matters. Zed's editor source is published under GPL-3.0-or-later; the others are proprietary.

Where identity and security fit

Each IDE holds credentials on the developer's machine: model provider keys (Zed stores saved keys in the system keychain), code host tokens, MCP server credentials and, for cloud agents, repository access. Treat them as non-human identities with an owner, scope and rotation date. See AI Coding Tools and Secrets Sprawl, and the tool security guides: Cursor Security Guide, Windsurf Security Guide and Devin Security for Devin Desktop, and Amazon Q Developer and Kiro Security Guide for Kiro. Extension marketplaces are also a supply chain: where a tool documents an extension allowlist (Cursor MDM policies, Devin Desktop AllowedExtensions), add it to the baseline, pin versions for extensions with broad access, and stage updates before wide rollout. See Browser Extension Security Guide for the same pattern in browsers.

Practitioner checklist

  • Record the vendor, tier and price checked, with the date, for every IDE on the shortlist.
  • Run one identical pilot task per IDE and log the usage consumed in the vendor's own unit.
  • Turn on the admin model allowlist and confirm which models are off by default.
  • Set terminal execution, web access and MCP to ask or allowlist mode for the pilot group.
  • Confirm the training and retention setting for the exact tier and enforce it centrally.
  • Enforce an extension allowlist through MDM or device policy where the tool supports it.
  • Inventory model keys, code host tokens and MCP credentials held by each IDE and assign owners.
  • Block personal accounts on the same tool once a company tier is in place.
  • Re-check pricing and model pages before renewal; lists on all five change monthly.

Standards and references

Related NHI Mgmt Group resources: AI Coding Tools Hub · Vibe Coding Tools Compared · AI Agent Platforms Compared · AI IDE and Coding Agent Security Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 15 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.