Guide · updated 2026-10-12 · 8 min read

Claude Code vs Codex vs Cursor (October 2026)

Claude Code, OpenAI Codex and Cursor can all read your repo, edit files, run commands and work in the cloud, so the real difference is where you work and how much you want to configure. Pick Claude Code if you live in the terminal and want the deepest configuration (CLAUDE.md, hooks, skills, subagents), Codex if you already pay for ChatGPT and want an agent bundled into it, and Cursor if you want an AI-first editor with tab completion and agents in one place.

Checked on 2026-10-12. All three ship changes weekly and pricing moves often. Every fact below links to the vendor's own docs or pricing page as of that date. If you're reading this months later, click through before you decide.

The short version

Sources for each row are linked in the sections below.

Claude Code

Claude Code is Anthropic's agentic coding tool. It started in the terminal and now also runs in VS Code and JetBrains, a desktop app, the browser and mobile. The CLI installs with curl -fsSL https://claude.ai/install.sh | bash or brew install --cask claude-code.

Where it's strongest: configuration. Claude Code has the most complete set of extension points of the three, and the best documentation for them:

• CLAUDE.md at user, project and local scope, plus path-scoped rules in .claude/rules/. See how to write a CLAUDE.md file.
• Skills with invocation control, arguments, dynamic shell context and forked execution. See our skills guide.
• Subagents with per-agent tools, models, MCP servers, hooks, memory and worktree isolation. See the subagents guide.
• Hooks on more than 30 lifecycle events, as command, HTTP, MCP tool, prompt or agent handlers. See hooks examples.
• MCP via claude mcp add, plus plugins that bundle all of the above.

It runs headless with claude -p, and in the cloud through Claude Code on the web, routines that run on a schedule or GitHub events, and GitHub Actions that respond to @claude mentions.

Pricing. Per claude.com/pricing on 2026-10-12, Claude Code is included in Pro ($20/month billed monthly, or $17/month billed annually), Max (from $100/month, at 5x or 20x Pro usage), Team (Standard seats $25/month monthly or $20 annually; Premium seats $125 monthly or $100 annually) and Enterprise. It is not in the Free plan. You can also pay per token through the API. Anthropic's cost docs put the average enterprise cost at around $13 per developer per active day.

Trade-offs. The configuration surface is large, and it's easy to over-build. Subscription usage is metered in rolling windows, and heavy agent use can hit limits on lower plans. There's no tab completion; it's an agent, not an editor.

OpenAI Codex

Codex is OpenAI's coding agent, available as a CLI, a desktop and web app, an IDE extension, and in the cloud. Install the CLI with the script at https://chatgpt.com/codex/install.sh, npm install -g @openai/codex or brew install --cask codex, then run codex and sign in with ChatGPT.

Where it's strongest: bundling and the cloud. If your team already has ChatGPT seats, Codex comes with them. Codex Cloud runs each task in its own workspace and lets you review results and open a pull request. You can kick off cloud tasks from the terminal with codex cloud. The IDE extension works in VS Code and compatible editors including Cursor and Windsurf, with separate integrations for JetBrains and Xcode.

Configuration. Codex has caught up on extension points:

• AGENTS.md from ~/.codex and from the repo root down to your working directory, with AGENTS.override.md support and a combined size cap (project_doc_max_bytes, 32 KiB by default).
• Skills in .agents/skills/ and ~/.agents/skills, invoked with /skills or $ in the CLI.
• Subagents defined as TOML files in .codex/agents/ or ~/.codex/agents/, with name, description and developer_instructions required.
• Hooks in hooks.json or config.toml, enabled by default, with events including PreToolUse, PostToolUse, Stop and SessionStart.
• MCP via codex mcp, and codex exec for scripts and CI.

Pricing. Per the Codex pricing page on 2026-10-12: Free and Go ($8/month) get limited access in the app; Plus ($20/month) adds the CLI, IDE extension, web and cloud integrations; Pro runs $100, $200 or $500/month; Business is $25/user/month monthly or $20 annually. Usage limits vary by model and task, and you can buy credits or use an API key at API rates (API key usage doesn't include cloud features).

Trade-offs. Usage limits are expressed as estimated message ranges per five-hour window, which makes capacity harder to plan. I find the configuration docs thinner than Anthropic's, and some features differ by surface.

Cursor

Cursor is an AI code editor first. Its agent works inside the editor, and it also has a CLI (installed with curl https://cursor.com/install -fsS | bash and run as agent) with Agent, Plan and Ask modes and -p for scripts.

Where it's strongest: the editor. If you want unlimited tab completion, inline edits and an agent in one window, Cursor is the most integrated option. Cloud Agents run in isolated cloud VMs and can be started from the web at cursor.com/agents, the desktop app, iOS, Slack, Linear, GitHub or Bitbucket comments, or an API. In the CLI you can hand a task to a Cloud Agent by prefixing your message with &.

Configuration.

• Rules as .mdc files in .cursor/rules/ with description, globs and alwaysApply frontmatter, plus AGENTS.md, user rules and team rules.
• Skills in .cursor/skills/ and .agents/skills/. Cursor's docs say it also reads .claude/skills/ and .codex/skills/.
• Hooks in .cursor/hooks.json or ~/.cursor/hooks.json. Cursor says it can load hooks from Claude Code, and exit code 2 blocks an action "for compatibility."
• MCP servers, and plugins.

Pricing. Per Cursor's pricing docs on 2026-10-12: Pro $20/month, Pro Plus $60/month, Ultra $200/month; Teams Standard seats $40/user/month and Premium $120/user/month. Usage beyond what's included is billed at model API prices. Cursor's own guidance is that daily agent users typically spend $60 to $100 a month and power users often spend $200 or more. Cloud Agents bill at API pricing and require a spend limit.

Trade-offs. Usage-based billing on top of the plan makes monthly cost less predictable. If you don't want to change editors, much of the value is in the editor itself.

How to choose

• You live in the terminal and want control. Claude Code. Hooks and subagents are the most mature of the three.
• Your company already pays for ChatGPT. Start with Codex; it costs you nothing extra to try.
• You want completion and an agent in your editor. Cursor.
• You want parallel cloud tasks you can start from your phone. All three do this now. Try the one bundled with a plan you already have.

You don't have to pick only one. A shared AGENTS.md, imported into CLAUDE.md with @AGENTS.md, plus skills in the Agent Skills format, gets you most of the way to a setup that works across all three.

Bottom line

As of October 2026 these tools have converged on the same building blocks: an instruction file, skills, subagents, hooks, MCP and cloud agents. The differences are surface and billing. Claude Code is the best-documented and most configurable, Codex is the easiest to justify if you're already in ChatGPT, and Cursor is the best if your editor is the center of your day. Whichever you choose, the habits in Claude Code best practices mostly transfer. Recheck prices before you commit; we'll update this page when they change.

Subscribe to Agent Shipped and we'll tell you when any of these change in a way that matters, twice a week.

FAQ

Which is better, Claude Code, Codex or Cursor?

None wins outright. Claude Code is strongest if you live in the terminal and want deep configuration through CLAUDE.md, hooks, skills and subagents. Codex fits teams already paying for ChatGPT. Cursor is the best fit if you want an AI-first editor with tab completion plus agents.

How much do they cost?

As of 2026-10-12, Claude Code is included with Claude Pro at $20 a month billed monthly, Codex is included with ChatGPT Plus at $20 a month, and Cursor Pro is $20 a month. All three have higher tiers and usage limits, and prices change often, so check the official pricing pages.

Can I use the same instruction file for all three?

Mostly. Codex and Cursor read AGENTS.md, and Claude Code v2.1.277 or later reads AGENTS.md when there is no CLAUDE.md. A CLAUDE.md containing @AGENTS.md lets you keep one shared file plus Claude-specific extras.

Can I use more than one of these tools?

Yes, and plenty of people do. Shared AGENTS.md instructions and the Agent Skills format make it cheaper to switch, and Cursor's docs say it reads .claude/skills/ and can load Claude Code hooks.