Skip to main content

The full matrix · 2026

All 9 AI coding tools, side by side

Every major AI coding tool in 2026, compared across the criteria that actually matter for picking one: interface, model, pricing, agent depth, MCP + skills support, open source, voice mode.

ToolVendorLaunchedInterfaceDefault modelPricingAgent modeSub-agentsScheduledAutocompleteMCPSkillsOpen sourceVoice
Claude CodeAnthropicOct 2024Terminal + IDE plugins (VS Code, JetBrains, Zed)Claude Opus 4.7 / Sonnet 4.6Bundled with Claude Pro $20/mo or Max $100, $200/moBest in 2026, sub-agents + scheduled tasks✓ Yes✓ YesLimitedFirst-classNativeProprietary✓ Yes
CursorAnysphere2023VS Code forkClaude, GPT, Gemini (selectable)$20/mo ProComposer + agent mode (mature), No, NoBest-in-classNativeVia skills-hub CLIProprietary, No
Antigravity 2.0GoogleMay 2026 (I/O)Desktop app + Go CLI + SDKGemini 3.5 FlashFree tier + Google AI subscriptionNative multi-agent architecture✓ YesVia skillsStrongNativeVia MCPSDK only✓ Yes
GitHub CopilotGitHub / Microsoft2021 (Copilot)VS Code / JetBrains / Visual Studio / Neovim extensionGPT, Claude, Gemini (selectable)$10/mo individual, $19+/user team, $39/user EnterpriseCopilot Workspace (beta), No, NoBest-in-classNativeVia MCPProprietary, No
WindsurfCodeiumJun 2025VS Code forkClaude, GPT, Codeium models$15/mo ProCascade flowsVia Cascade, NoStrongNativeVia MCPProprietary, No
Codex CLIOpenAIMar 2025 (open-source)Terminal CLIGPT-5 / GPT-5.5Bundled with ChatGPT Pro/Plus $20, $200/moStrong terminal agent, No, NoLimitedNativeNative✓ Open, No
ClineCline Bot Inc.Sep 2025 (1.0)VS Code extensionAny (BYO API key)Free (BYO API key)Per-step approval agent, No, NoLimitedNativeVia MCP✓ Open, No
AWS KiroAWS / Amazon BedrockJan 2026VS Code-based IDEAmazon Bedrock (Claude, Llama, Titan)Pay-as-you-go via AWS BedrockSpec-driven loopVia Cascade, NoStrongVia Bedrock AgentCoreVia MCPSDK only, No
Gemini CLIGoogle2025Terminal CLIGemini 2.5 Pro / 3.5 FlashFree for individual useStrong terminal agent, No, NoLimitedNativeVia MCP✓ Open, No

How to read this matrix

  • Sub-agents = native parallel isolated agents with their own context windows. Currently only Claude Code and Antigravity 2.0 have first-class support.
  • Scheduled = cron-style background runs. Only Claude Code's /loop is native; others require external cron or skills.
  • Autocomplete = inline tab completion quality. "Best" = best-in-class; "Strong" = comparable to leaders; "Limited" = not the tool's focus.
  • MCP = Model Context Protocol support quality. First-class = native + first-party servers; Native = full client support; Via Bedrock = routed through AWS AgentCore.
  • Skills = SKILL.md (Agent Skills standard) support. Native = recognized as a first-class concept; Via MCP = available through MCP prompts; Via skills-hub CLI = installs via npx into the right per-tool directory.

Need a recommendation?

5-question quiz that recommends the right tool for your stack and workflow.

Take the quiz →

Want head-to-head?

31 pairwise comparison pages, Claude Code vs Cursor, Cursor vs Windsurf, Antigravity vs Kiro, and more.

All comparisons →

One skills catalog, all 9 tools

Skills follow the open Agent Skills standard. Install once, use in any tool above.

Browse 4,900+ skills →