Codex vs Claude Code for Solo Founders (2026)

Verdict: Pick Codex if you already pay for ChatGPT, want a CLI you can fork under Apache-2.0, or want local commands to run inside a workspace sandbox by default. Pick Claude Code if you already pay for Claude, prefer its long-session workflow (our opinion, not a benchmark), or want an agent that starts in a classifier-reviewed auto mode in the terminal. If you can afford $40/month, run one as the builder and the other as the reviewer.

Every price, plan name, limit and default below comes from OpenAI's, Anthropic's or the project's own pages, checked on 15 Nov 2026. The judgement calls are ours and marked as such.

TL;DR#

• Entry price is a tie. ChatGPT Plus and Claude Pro are both $20/month, and both include their coding agent (Codex pricing, Claude pricing).
• Codex scales higher. Pro comes in $100, $200 and $500 tiers, and Ultrafast is listed for Pro $500 and eligible Enterprise and Edu plans (Codex pricing). Claude Max is $100 and Ultra is $200 (Claude pricing).
• Codex is the open one. The Codex CLI is Apache-2.0, while OpenAI says the IDE extension and cloud are not open source (openai/codex, Codex open source). Claude Code's repository licence reads "all rights reserved" (LICENSE.md).
• Instructions differ. Codex reads global and nested AGENTS.md files, capped at 32 KiB by default (Codex AGENTS.md). Claude Code reads CLAUDE.md, and by default reads AGENTS.md only when no CLAUDE.md is present (Claude memory).
• Both are serious CI tools. Codex has codex exec in a read-only sandbox by default. Claude Code has claude -p with budget and turn caps, plus a GitHub Action (Codex non-interactive, Claude CLI reference).

Disclosure: ZeroRaised publishes this comparison and is a launch site for bootstrapped products. ZeroRaised has no commercial relationship with OpenAI or Anthropic and earns nothing if you subscribe to either. Facts come from the linked sources as checked on 15 Nov 2026.

What this comparison is not#

We have already covered the editor question in Claude Code vs Cursor for solo founders, the wider editor market in Cursor alternatives for indie developers, and the review process in how to review AI-generated code. This post is narrower: two terminal-first agents and the plans that include them.

Plans and prices#

OpenAI says Codex is included in ChatGPT Free, Go, Plus, Pro, Business and Enterprise, with "Expanded Codex usage" on Plus and "Maximum Codex tasks" on Pro (Codex plans, Codex pricing). Anthropic lists Claude Code from Pro upward and excludes it from Free (Claude pricing).

ChatGPT Pro is now offered at three price points ($100, $200 and $500), and Ultrafast is listed for Pro $500 and eligible Enterprise and Edu plans (Codex pricing). Annual billing also changes the Claude maths: Pro is $17/month when you pay $200 up front (Claude pricing).

Our read: if you already pay for one, the agent is free at the margin; adding the other costs a second $20 rather than saving the first.

Usage limits are not comparable#

Do not convert one vendor's message count into the other's. They count different things.

For Plus, OpenAI publishes a range per model, without stating the unit: GPT-6.1 Sol is 15 to 160, GPT-6 Astra 5 to 45, GPT-6 Sol 15 to 150, GPT-6 Luna 350 to 3,000, GPT-5.6 Sol 10 to 100 and GPT-5.6 Luna 250 to 2,000. Local messages and cloud chats share your plan's usage allowance (Codex plans).

Anthropic uses multipliers: Pro is the 1x baseline, Max is 5x Pro every five hours and every week, and Ultra is 10x Pro weekly with no five-hour limit for verified accounts (Claude pricing).

Those are different units, so compare a week of your own work rather than the raw numbers. OpenAI also lets Plus users extend usage with ChatGPT credits (Codex pricing).

Worked monthly cost example (API keys)#

API keys are how you use either agent in CI or your own orchestration. This is an illustrative example, not a quote: 30M input tokens and 3M output tokens in a month, no caching or batch discount, at standard list prices checked on 15 Nov 2026 (gpt-6.1-sol is the short-context, 272K or less, rate).

Sources: OpenAI API pricing, Claude API pricing.

The arithmetic is 30 x input price plus 3 x output price. That is why the two mid-tier models tie at $90. Cached input is cheaper and batch work is discounted, so a real bill is usually lower.

Our read: at this volume, a $20 subscription is far cheaper than API keys. Use the API route when you are automating runs you cannot do by hand.

Surfaces: CLI, IDE, cloud#

Codex runs in the terminal, as an IDE extension, in Codex cloud and in the desktop app for macOS and Windows (Codex open source, Codex plans). OpenAI says "Each task has its own workspace and can keep working while your computer is asleep," and you can inspect changed files and commit or open a pull request (Codex Cloud).

Claude Code runs in the terminal, VS Code, JetBrains, a desktop app and the browser, and sessions can move between them (Claude Code overview). Cloud sessions run in isolated Anthropic-managed VMs with network access limited by default (Claude security).

That is close to parity. Codex's cloud flow is built around a task you hand off; Claude Code's around moving one session between terminal, desktop and phone.

Instructions files: AGENTS.md vs CLAUDE.md#

Codex looks for a global file in ~/.codex, then walks from the project root to your current directory, concatenating files so that the ones closest to you come later and override earlier guidance. It stops adding files once the combined size reaches 32 KiB by default (Codex AGENTS.md).

Claude Code reads CLAUDE.md at the start of every session. It also supports CLAUDE.local.md, path-scoped rules and auto memory. By default, if your repo has AGENTS.md and no CLAUDE.md, Claude Code reads AGENTS.md instead, and a setting lets you make it read both (Claude memory).

Our read: put shared rules in AGENTS.md and add a short CLAUDE.md that imports it with @AGENTS.md. Anthropic documents the @AGENTS.md import, and you avoid maintaining the same architecture notes twice.

Sandbox and permission defaults#

This is the biggest behavioural difference for a solo founder, because it decides how often the agent stops to ask.

Codex applies sandboxing automatically in its default permissions mode. On macOS it uses Seatbelt, on Windows its native sandbox, and on Linux bubblewrap. The default sandbox mode is workspace-write, which OpenAI calls "the default low-friction mode for local work," and the docs pair it with the on-request approval policy (Codex sandboxing). In automation, codex exec runs in a read-only sandbox by default; you opt into edits with --sandbox workspace-write (Codex non-interactive).

Anthropic says auto mode is the built-in starting permission mode for interactive terminal and VS Code sessions, where a separate classifier model reviews actions and blocks the ones it judges unsafe. Manual mode starts with read-only permissions. A sandboxed Bash tool with filesystem and network isolation is available with /sandbox (Claude security).

Neither default is a guarantee. Both vendors say you remain responsible for reviewing what ships. Our read: Codex's default is easier to reason about as a boundary, while Claude Code's is easier to live with. Widen either last, not first.

Headless runs and CI#

Codex uses codex exec. It streams progress to stderr, prints the final message to stdout, supports --json for JSON Lines, --output-schema for structured output and resume for a second stage (Codex non-interactive). For GitHub Actions, OpenAI recommends its Codex GitHub Action instead of installing and authenticating the CLI yourself, because the action starts a secure proxy for the API key.

Claude Code uses claude -p, with JSON and stream-JSON output, --max-turns, --max-budget-usd (checked against a client-side cost estimate that can differ from your bill) and --json-schema (Claude CLI reference). Its GitHub Action answers @claude mentions and can run from a prompt on any GitHub event, including a schedule (Claude GitHub Actions).

Our read: for a scheduled job that should look but not touch, Codex's read-only default is safer. For a @claude mention with a per-run turn or estimated-spend cap, Claude Code's action is more complete.

GitHub review#

Codex reviews pull requests when you comment @codex review, and can run automatic reviews. OpenAI says that in GitHub it flags only P0 and P1 issues, and you can add a ## Code Review Rules section to AGENTS.md (Codex GitHub).

Claude Code's GitHub Action can post a review as inline comments, skips draft and closed pull requests and ones it judges trivial, and can run the code-review plugin (Claude GitHub Actions).

For a deeper comparison of review tools rather than coding agents, see our best AI code review tools for solo developers. Our read: both are good second pairs of eyes, and neither replaces tests or branch protection.

MCP and tool connections#

Both agents speak the Model Context Protocol, so the same server can usually serve both.

Codex stores MCP configuration in config.toml, supports STDIO and Streamable HTTP servers, OAuth and bearer tokens, tool allow lists and per-tool approvals (Codex MCP). Claude Code supports MCP servers, and project-scoped servers live in a .mcp.json you can check into source control (Claude Code overview, Claude security).

Our read: one MCP setup for your tracker, database or docs usually works for both. That is one fewer lock-in than the prices suggest.

Open source and lock-in#

The Codex CLI is Apache-2.0, and OpenAI lists the SDK and app server as hosted in the same repository (openai/codex, Codex open source). OpenAI is explicit that the IDE extension and Codex cloud are "Not open source" (Codex open source).

Claude Code's repository is public, but its licence is not open source. It reads "© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service" (LICENSE.md).

Our read: if licence terms matter, for example because you want to audit the client, Codex is the only real option. If you only care about the agent, both are hosted services and the licence changes little.

Picks by situation#

The use-both workflow#

If you bill enough to afford $40/month, this is the setup we would run:

1. Write the spec once. Put architecture notes, test commands and review rules in AGENTS.md. Add a CLAUDE.md that imports it.
2. Build with one. Use the subscription you already have, and keep changes small enough to review in one sitting.
3. Review with the other. Ask the second agent to review the diff against the spec, not rewrite it. A different vendor catches different mistakes.
4. Automate the boring part. Schedule codex exec or claude -p for dependency updates, changelog drafts or flaky-test triage, with read-only permissions and a spending cap.
5. Ship. Once tests pass, follow our product launch checklist for bootstrapped makers and launch your app.

Our read: the second subscription is worth more as a reviewer than as a second builder. Two agents on the same branch is a merge conflict with extra steps.

Frequently asked questions#

Is Codex free for solo founders?#

There is a free tier, but OpenAI describes it as limited trial access to Codex. The practical entry point is ChatGPT Plus at $20/month, which OpenAI lists with expanded Codex usage and describes as powering a few focused coding sessions each week. Codex is also included in Go, Pro, Business and Enterprise plans.

Is Claude Code included in Claude Pro?#

Yes. Claude Code is included in Pro at $20/month billed monthly, or $17/month with the annual plan paid as $200 up front. It is also included in Max at $100/month, Ultra at $200/month, Team and Enterprise. The Free plan does not include Claude Code.

Which is cheaper for a solo founder?#

At entry level they cost the same: ChatGPT Plus and Claude Pro are both $20/month. If you pay per token through an API key, list prices are close for mid-tier models. In our worked example, 30M input and 3M output tokens cost $94.50 on gpt-5.3-codex, $90 on gpt-6.1-sol and $90 on Claude Sonnet 5.5, while Claude Opus 5.5 came to $180.

Can I use Codex and Claude Code together?#

Yes, and it is a sensible setup. Both work from the same repository and read instruction files, so you can let one agent implement and the other review the diff. The cost is two subscriptions, from $40/month at the entry tiers, or API usage if you prefer pay-as-you-go.

Do I need to be a developer to use either?#

You need to be able to read a diff and run your project, because both agents make real edits. In our view a non-technical founder is better served by Codex's desktop app and cloud tasks, where OpenAI says you can inspect changed files and open a pull request without living in a terminal. Neither tool removes the need to check what shipped.

Build, then launch it#

ZeroRaised is a launch directory for products built without institutional VC. When your agent-assisted build is ready for users, submit your product for free. For more comparisons and guides on shipping without investors, join the ZeroRaised newsletter.

Sources#

All sources last checked 15 Nov 2026. OpenAI's help-centre articles returned HTTP 403 when we checked, so no claim relies on them; OpenAI's developers.openai.com docs now redirect to learn.chatgpt.com, which is where we read them.

• Codex plans, last checked 15 Nov 2026
• Codex pricing docs, last checked 15 Nov 2026
• OpenAI API pricing, last checked 15 Nov 2026
• Codex sandboxing, last checked 15 Nov 2026
• Codex non-interactive mode, last checked 15 Nov 2026
• Codex AGENTS.md, last checked 15 Nov 2026
• Codex Cloud, last checked 15 Nov 2026
• Codex GitHub review, last checked 15 Nov 2026
• Codex MCP, last checked 15 Nov 2026
• Codex open source, last checked 15 Nov 2026
• openai/codex on GitHub, last checked 15 Nov 2026
• Claude pricing, last checked 15 Nov 2026
• Claude API pricing, last checked 15 Nov 2026
• Claude Code overview, last checked 15 Nov 2026
• Claude Code CLI reference, last checked 15 Nov 2026
• Claude Code security, last checked 15 Nov 2026
• Claude Code memory, last checked 15 Nov 2026
• Claude Code GitHub Actions, last checked 15 Nov 2026
• Claude Code licence, last checked 15 Nov 2026