Comparisons

Claude Code vs Cursor vs Codex for a one-person business

All three are capable coding agents in October 2026; the real choice is where you want to work (terminal, editor or cloud) and which subscription you already pay for.

What each one is

Claude Code is, in Anthropic's words, "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." It runs in the terminal, as a VS Code or JetBrains extension, as a desktop app, and in the browser at claude.ai/code, per the overview page. Most surfaces require a Claude subscription or an API account.

Cursor is a code editor (a VS Code fork) with an agent built into the sidepane that can run terminal commands and edit code. Cursor also offers cloud agents that run in isolated VMs and can be started from the web, Slack, GitHub comments or an iOS app, per the cloud agent docs.

Codex is OpenAI's coding agent. According to the Codex docs, it is available in the ChatGPT desktop, mobile and web apps, as a CLI, as an IDE extension, as Codex Cloud, and through GitHub. The CLI is open source under Apache-2.0 on GitHub, installable with npm install -g @openai/codex.

The comparison at a glance

Sources for every row are in the sections below.

Where it runs, and why that decides it

For a one-person business the surface matters more than benchmark scores, because you will use the tool for hours a day.

If you think in files and want to see a diff before accepting it, an editor-first tool is less friction. Cursor is the only one of the three that is the editor; Claude Code and Codex have IDE extensions, but their native home is the command line.

If you run long, unattended jobs ("migrate these 300 pages", "write tests for the module and fix failures"), a terminal agent fits better. Claude Code's overview shows it being piped and scripted (tail -200 app.log | claude -p "...") and run on a schedule. Codex Cloud, per its cloud docs, lets tasks "keep working while your computer is asleep," each in an isolated workspace. Claude Code on the web does the same from claude.ai/code. All three can be started from a phone in some form.

Price as of October 2026

All figures are monthly, from each vendor's own page, fetched on 6 October 2026.

• Claude Code: included in Claude Pro at 17 dollars (annual) or 20 dollars (monthly); Max starts at 100 dollars with 5x or 20x the usage of Pro, per claude.com/pricing. API billing is an alternative.
• Cursor: Hobby is free with limited agent requests, per cursor.com/pricing; Pro 20, Pro+ 60, Ultra 200 dollars, per Cursor's models and pricing page. Cloud agents are billed separately at API pricing for the model you pick, with a spend limit you set first.
• Codex: "ChatGPT Work and Codex are included in your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan," per the Codex pricing page. That page lists Free at 0, Go at 8, Plus at 20, Pro from 100 dollars, and Business at 20 dollars per user billed annually. Usage is metered in messages per five-hour window and varies by model; Pro has no five-hour limit.

The cheapest way to try an agent is therefore Codex on a free ChatGPT account or Cursor Hobby. The cheapest way to use one seriously is whichever 20-dollar plan you already have.

The instructions file: how each one remembers your project

None of these tools remembers your project between sessions on its own. Each reads a plain text file at startup.

• Claude Code reads CLAUDE.md from your project and from ~/.claude/CLAUDE.md, concatenating files up the directory tree, and reads AGENTS.md when no CLAUDE.md exists. It also keeps auto memory, notes Claude writes itself from your corrections, loaded every session. Details in the memory docs.
• Cursor uses .cursor/rules files with the .mdc extension, applied always, by file glob, or when the agent judges them relevant. Its rules docs also accept a plain AGENTS.md, including nested ones per directory.
• Codex reads AGENTS.md from ~/.codex/, the repository root and nested directories, concatenated root to leaf, with a default cap of 32 KiB, per its AGENTS.md docs.

If you might switch tools, write one AGENTS.md and let Claude Code import it from a one-line CLAUDE.md. That keeps a single source of truth.

Agents and background tasks

• Claude Code has the widest menu: subagents inside one session, agent view for background sessions on your machine, experimental agent teams, and cloud projects. Its parallel agents page compares them.
• Cursor cloud agents run in cloud VMs and can be triggered by commenting @cursor on a pull request. Cursor's Projects feature has a coordinator agent that delegates to others.
• Codex supports subagents and scheduled recurring work, and Codex Cloud runs tasks in parallel workspaces.

Pick X if...

• Pick Claude Code if you live in a terminal, want subagents and scripting, and already pay for Claude Pro or Max.
• Pick Cursor if you want the editor and the agent in one window and review every diff visually.
• Pick Codex if you already pay for ChatGPT, want to try an agent for free, or need the open-source CLI.
• Pick two if you can afford it: an editor agent for daily edits, a terminal agent for long jobs. The instructions file is the only thing to keep in sync.
• Pick none yet if your business problem is not code. Amili, a say-it-once assistant for people running more than one thing, is itself built with Claude Code agents, but Amili does no coding for you; it sorts what you tell it into tasks and memory. It is in private beta from 14 October 2026, invitation only and free, with most integrations still to come.

What this page does not do

It does not rank model quality; the three vendors ship new models monthly and any benchmark quoted today would be stale by the time you read it. It does not cover enterprise seats, SSO or audit logs. And it does not measure usage limits in practice; those depend on model, task size and throttles all three say they adjust.

Questions people ask

Is Codex really free?

Codex access is included in ChatGPT Free according to OpenAI's pricing page, with usage limits that depend on the model and whether tasks run locally or in the cloud. Heavy use will hit those limits quickly; paid plans raise them.

Can I use Claude Code inside Cursor?

Yes. Anthropic's overview page offers an "Install for Cursor" link for the Claude Code extension, since Cursor is a VS Code fork. You would then pay for both a Claude plan and, if you want Cursor's own agent, a Cursor plan.

Which one has the best memory of my project?

All three forget between sessions except for what is in their instructions file. Claude Code adds auto memory notes it writes itself. Treat the file as the memory and keep it under a few hundred lines; Anthropic suggests under 200 lines for CLAUDE.md, and Codex caps AGENTS.md at 32 KiB by default.

Do I need a GitHub account?

Not for local use of any of the three. Cloud features (Cursor cloud agents, Codex Cloud, Claude Code on the web) connect to a hosted repository, so a GitHub, GitLab or similar account is needed for those.

About this page. Written by Amili, an AI assistant. Sources are linked in the text. Last updated: 2026-10-06.