Where they actually differ
Codex is OpenAI’s coding agent; Claude Code is Anthropic’s. As of November 2026 they look more alike than ever: both run in the terminal, in the IDE, in a desktop app and in the cloud, both connect to MCP servers, and both take direction from an instruction file in the repo. The differences are in the details.
| Aspect | Codex | Claude Code |
|---|---|---|
| Where it runs | ChatGPT desktop app, Codex CLI, Codex IDE extension, Codex Cloud, web and iOS (from Plus) | Terminal, VS Code and Cursor, JetBrains, desktop app, web and mobile |
| Team plan | ChatGPT Business: $20 per user per month billed annually, $25 monthly, 2+ users | Claude Team: standard seat $20 per seat per month billed annually, $25 monthly, 2 to 150 seats |
| More usage | Buy workspace credits | Premium seat: $100 per month billed annually ($125 monthly), 5x usage |
| Instruction file | AGENTS.md (and AGENTS.override.md) | CLAUDE.md; from v2.1.277 reads AGENTS.md when there is no CLAUDE.md or CLAUDE.local.md |
| Memory | Codex memories, off by default, local under ~/.codex | Auto memory, on by default, local under ~/.claude/projects/<project>/memory/ |
| Cloud tasks | Codex Cloud: each task gets its own workspace, with or without a repo from a connected GitHub or GitLab account | Claude Code on the web and Cloud scheduled tasks |
Prices as of November 2026, from the OpenAI and Anthropic pricing pages.
How a team should choose
- Check what you already pay for. If the team has ChatGPT Business, Codex is included. If it has Claude Team, Claude Code is included. A second subscription on top costs at least $20 more per person per month.
- Have two developers spend a week on a real task with each. Benchmarks on blogs vary by model generation and are often written by parties with a stake. Your own repo is the only test that counts.
- Look at the cloud side. If the team wants tasks that keep running with the laptop closed, and reviews on every pull request, both offer that. Codex Cloud can start without a repo, or check out repos from a connected GitHub or GitLab account. Codex on an API key alone has no cloud features such as GitHub review or Slack.
- Look at administration. ChatGPT Business gives a dedicated workspace with admin controls, SAML SSO and MFA. SCIM, audit logs via the Compliance API and retention controls come with Codex Enterprise. Hold your IT requirements against both pricing pages before you decide.
The misconception: you have to pick one
Most teams assume they must standardize on one tool, because otherwise the instructions get written twice. That is no longer true. AGENTS.md is an open format read by Codex, Cursor, GitHub Copilot coding agent, Gemini CLI and dozens of other tools. Claude Code reads it too: from v2.1.277 directly when there is no CLAUDE.md or CLAUDE.local.md, and through an @AGENTS.md line when there is one. One file in the repo, and every developer keeps their own tool. The exact mechanics are in What is AGENTS.md and how does Codex read it?.
What does not get shared is what the tools learn on their own. Codex memories live on the machine of whoever turned them on. Claude Code’s docs say it plainly: “Auto memory is machine-local.” What Claude Code learned about your test suite on one laptop, Codex on another laptop does not know, and neither does Claude Code on a colleague’s machine. Switch tools and that memory starts over. More in Does Codex remember previous work?.
What this means for a team
The choice between Codex and Claude Code is smaller than it looks. Seat prices match, the surfaces are comparable, and instructions can be shared. The real question is where the team’s knowledge lives: in an AGENTS.md someone maintains, in scattered memories on laptops, or in one place every tool reads. What Codex costs a team in detail, with limits and credits, is in How much does Codex cost for a team?. How to capture company knowledge for AI at all is in How to give AI agents your company context.
Where Bombos fits
Bombos does not pick between Codex and Claude Code: it is a plugin that gives both, and the team’s other AI tools, one shared memory. What one developer learns with Codex and the rules the team agrees on are then ready for the colleague using Claude Code, traceable, in the cloud or on your own server. One developer can try it first for their own agents.
Related questions
Can we use Codex and Claude Code side by side in one repo?
Yes. Put project conventions in AGENTS.md. Codex reads it automatically. Claude Code v2.1.277 or later reads it when there is no CLAUDE.md or CLAUDE.local.md in or above the working directory, or through an @AGENTS.md line inside a CLAUDE.md.
Which one is cheaper for a team?
Per seat they match: $20 per user per month billed annually, $25 monthly (November 2026). The difference is in usage limits and top-ups: Codex uses credits, Claude Team offers a premium seat at $100 per month billed annually.
Do Codex and Claude Code keep working when my laptop is off?
Both do. Codex Cloud gives each task its own workspace that keeps going while your computer sleeps. Claude Code has Claude Code on the web and Cloud scheduled tasks that run in the cloud.
Do Codex and Claude Code share what they learned about our project?
No. Codex keeps memories under ~/.codex and Claude Code keeps auto memory under ~/.claude/projects. Both live on one developer's machine, and neither tool picks up the other's memories automatically.
