Cursor vs Antigravity vs Claude Code: how they compare
Cursor vs Google Antigravity vs Claude Code from their own docs: surfaces, models, instruction files, skills, cloud agents, and which fits your team.
A team lead is asked to standardize on one AI coding tool. Half the team uses Cursor, two people switched to Antigravity after Gemini CLI changed, and the platform team runs Claude Code in CI. Every comparison they find is either a benchmark that will be out of date next month or written by one of the vendors.
This guide compares the three from their own documentation, checked on 16 November 2026: Cursor, Google Antigravity, and Claude Code. It leaves out prices and benchmarks, which change too fast to be honest about.
The short answer
| Cursor | Antigravity | Claude Code | |
|---|---|---|---|
| Center of gravity | The editor | The agent manager | The agent, wherever you run it |
| Surfaces | Editor, CLI, web, iOS | Desktop app, CLI (agy), extensions for 6 IDEs, Python SDK | Terminal, VS Code and JetBrains, desktop app, web, mobile |
| Models | Several providers: Claude, Gemini, GPT, and Grok models | Gemini 4 family, plus Claude and GPT-OSS depending on plan | Claude |
| Instruction files | Rules, AGENTS.md | AGENTS.md or GEMINI.md, .agents/rules/ | CLAUDE.md, .claude/rules/ |
| Skills | Agent Skills format | Agent Skills format | Agent Skills format |
| MCP, hooks, subagents | Yes | Yes | Yes |
| Agents in the cloud | Cloud Agents | Remote Agents | Cloud sessions |
| Programmatic use | Agent SDK | Antigravity SDK (Python) | Agent SDK (Python, TypeScript), claude -p |
Cursor
Cursor is an AI code editor, with the agent built into the editing experience. Its docs cover rules, skills, hooks, subagents, and MCP, and its model list spans Claude, Gemini, GPT, and Grok models.
Two things stand out:
- Cloud Agents run in isolated VMs with full development environments. You can run many in parallel, start them from the web or the iOS app, and they can build, test, and open pull requests, including across multiple repositories.
- Model choice. One tool, several providers, so teams can switch models per task without switching tools.
Fits developers who want an AI-first editor and model flexibility. See Cursor rules vs skills vs AGENTS.md vs CLAUDE.md.
Google Antigravity
Antigravity is Google's agent platform. Its docs describe four surfaces: the Antigravity App, a desktop command center that evolved from the earlier Agent Manager, for launching and inspecting agents with plan and diff review; the Antigravity CLI (agy); extensions for six IDEs; and a Python SDK.
What stands out:
- Gemini 4 models, including an Orchestrator option that pairs a larger planner model with the faster Gemini 4 Hydrogen for implementation. Claude Sonnet 5.5 and Opus 5.5 are listed for Google AI Pro, Ultra, and Enterprise.
- Rules with triggers.
AGENTS.mdorGEMINI.md, plus.agents/rules/*.mdfiles that load always, by glob, manually, or when the model decides they're relevant. - Plan-dependent features. Which models you get depends on your Google plan; check the model table before deciding.
Fits teams on Google Cloud, people moving from Gemini CLI, and anyone who wants a desktop view of several agents at once. See Gemini CLI vs Claude Code vs Codex CLI.
Claude Code
Claude Code is Anthropic's coding agent. It runs in the terminal, in VS Code and JetBrains, in the Code tab of the Claude desktop app, and as cloud sessions on the web and mobile.
What stands out:
- Configuration depth.
CLAUDE.md, skills, hooks at many lifecycle points, permission modes including auto mode with a safety classifier, and managed settings that let an organization enforce policy. - Scaling work. Subagents, experimental agent teams, and dynamic workflows that orchestrate many subagents from a script.
- Automation.
claude -p, GitHub Actions, and the Agent SDK use the same agent.
Fits terminal-first developers and teams that want one configured agent across every surface. See Claude Code best practices.
Running more than one
Many teams do, and the shared formats make it cheaper:
AGENTS.mdas the shared instruction file. Cursor and Antigravity read it; Claude Code imports it fromCLAUDE.mdwith@AGENTS.md.- One skills folder. Keep skills in
.claude/skills/. Cursor reads that folder for compatibility; symlink.agents/skillsto it for Antigravity. See the Agent Skills open standard. - MCP servers configured in each tool, pointing at the same servers.
How to choose
- You live in your editor and want model choice: Cursor.
- You're on Google Cloud, want Gemini 4, or like a desktop agent manager: Antigravity.
- You want the terminal, deep team configuration, or the same agent in CI: Claude Code.
- Still unsure: give two of them the same real task in your own repository for a week. That tells you more than any comparison.
What none of them knows
All three read your code and your instruction files. None of them knows what the team decided in this week's planning meeting unless someone writes it where the agent looks, and keeps it current.
Where Lockstep fits
Lockstep captures decisions from PRDs, Slack, and pull requests, routes each to its owner for approval, and briefs Claude Code sessions automatically on the ones that apply. Session briefings, hooks, and managed skills are Claude Code features; Cursor, Antigravity, and other MCP clients can call Lockstep's tools directly.
See how it works.
Sources checked 16 November 2026
- https://cursor.com/docs/cloud-agent
- https://cursor.com/llms.txt
- https://antigravity.google/docs/overview
- https://antigravity.google/docs/models
- https://antigravity.google/docs/rules
- https://code.claude.com/docs/en/overview
Related reading
For Claude Code against Cursor and Codex, see Claude Code vs Cursor vs Codex. For Claude Code's own surfaces, see Claude Code desktop vs CLI vs VS Code. For Kiro, see Kiro vs Claude Code. For Cursor against GitHub Copilot, see Cursor vs GitHub Copilot.
Frequently asked questions
- What is Google Antigravity?
- Google's agentic coding platform. Its docs list four surfaces: the Antigravity App, a desktop command center for running and inspecting agents; the Antigravity CLI, agy; extensions for six IDEs; and a Python SDK. It replaced Gemini CLI for individual users in June 2026.
- What is the difference between Cursor and Antigravity?
- Cursor is built around its code editor, with agents inside it, plus cloud agents and a CLI. Antigravity is built around agents, with a desktop app for orchestrating them and extensions that bring them into existing IDEs. Cursor offers models from several providers; Antigravity centers on Gemini 4 and also lists Claude models on some plans.
- Can I use Claude models in Cursor or Antigravity?
- Yes, in both. Cursor's docs list Claude Sonnet 5.5, Opus 5.5, and other Claude models alongside Gemini, GPT, and Grok models. Antigravity's model table lists Claude Sonnet 5.5 and Opus 5.5 on Google AI Pro, Ultra, and Enterprise plans. Claude Code runs Claude models.
- Do Cursor, Antigravity, and Claude Code share instruction files?
- Partly. Cursor and Antigravity read AGENTS.md; Antigravity also reads GEMINI.md, and Claude Code reads CLAUDE.md, which can import AGENTS.md. All three support SKILL.md skills in the Agent Skills format, from different folders.
- Which is best: Cursor, Antigravity, or Claude Code?
- It depends on how you work. Cursor suits developers who want an AI-first editor. Antigravity suits teams on Google's stack or wanting Gemini 4 with a desktop agent manager. Claude Code suits terminal-first developers and teams that want deep configuration through hooks, settings, and managed policies.
Keep reading