What changed recently
Two things to know before you read older comparisons:
- GPT-5.5 is gone from Codex for ChatGPT users. It was retired on October 14, 2026 for Codex with ChatGPT sign-in. OpenAI now recommends GPT-6.1 Sol for complex coding (Plus and higher) and GPT-6 Luna for focused tasks and on the Free and Go plans. API-key sessions weren’t affected. (OpenAI’s Codex pricing page)
- Codex now lives inside ChatGPT. The desktop experience is a mode in the ChatGPT desktop app, and Codex shares usage limits and credits with ChatGPT Work. Pro now comes in three tiers: $100, $200 and $500.
The key difference
Claude Code and Codex are the most direct rivals in this space. Both are terminal-first coding agents from a model lab. Both are also available in IDE extensions, a desktop app and the cloud, and both are bundled into that lab’s chatbot subscription.
The differences that actually decide it:
- Models. Claude Code runs Claude only. Codex runs OpenAI’s GPT models on ChatGPT plans, and its CLI can be configured for local models (Ollama, LM Studio) and other OpenAI-compatible providers.
- Openness. The Codex CLI is open source under Apache-2.0: you can read, fork and audit it. Claude Code is closed source.
- Extensibility is close. Both support MCP servers, hooks, subagents and plugins, so this rarely decides it. Claude Code also has skills; Codex has rules and memories.
- Subscription. Claude Code comes with Claude plans; Codex comes with ChatGPT plans. If you already pay for one chatbot, that’s often the tie-breaker.
Pricing in practice
| Tier | Claude Code (Claude plans) | Codex (ChatGPT plans) |
|---|---|---|
| Free | Not available | Free: limited, GPT-6 Luna in the desktop app (subject to rollout) |
| $8 | — | Go: lightweight use, GPT-6 Luna in the desktop app |
| $20 | Pro: $20/month ($17/month billed yearly) | Plus: $20/month. Codex on web, CLI, IDE extension and iOS; cloud code review and Slack |
| $100 | Max 5x | Pro ($100 tier) |
| $200 | Max 20x | Pro ($200 tier) |
| $500 | — | Pro ($500 tier), with Ultrafast mode |
| Teams | Team: Standard seat $25/user ($20 billed yearly), Premium seat $125 ($100 billed yearly) | Business: $25/user ($20 billed yearly), 2+ users |
| Pay per token | Anthropic API key | OpenAI API key |
Prices are from the official Claude pricing and Codex pricing pages, checked October 22, 2026.
Two details that matter:
- Codex’s cloud features need a ChatGPT plan. Automatic code review on GitHub and the Slack integration come with Plus and up, not with an API key.
- Usage limits differ by model and task on both sides, so the plans don’t translate into a fixed number of tasks. OpenAI’s own pricing page warns that API token prices shouldn’t be used to estimate how many tasks a subscription includes.
Features side by side
| Claude Code | Codex | |
|---|---|---|
| Terminal agent | Yes (closed source) | Yes (Apache-2.0) |
| IDE | VS Code, Cursor, JetBrains | VS Code, Cursor, Windsurf, JetBrains, Xcode |
| Desktop | Claude Code desktop app | Codex mode in the ChatGPT desktop app |
| Cloud | Cloud sessions in the browser | Cloud agent that runs tasks and opens pull requests |
| Project instructions | CLAUDE.md (and AGENTS.md in supported setups) |
AGENTS.md (/init creates it) |
| Extensions | MCP, hooks, subagents, skills, plugins | MCP, hooks, subagents, plugins, rules |
| Models | Claude | GPT-6.1 Sol, GPT-6 Luna; local and custom providers in the CLI |
| Free tier | No | Limited, on ChatGPT Free and Go |
Who should pick which
Choose Claude Code if:
- you prefer Claude models, or already pay for Claude
- you want to run Claude through Amazon Bedrock, Google Cloud or Microsoft Foundry for an enterprise deployment
- you want one agent across terminal, IDE, desktop and browser on a Claude subscription
Choose Codex if:
- you already pay for ChatGPT, so Codex costs nothing extra
- you want an open-source CLI you can audit or fork, or want to point it at local models
- you want a free way to try a coding agent (ChatGPT Free and Go include limited use)
- you need Xcode or Windsurf integration
Use both if: you want a second opinion on your code. Teams commonly let one agent write a change and the other review it. Keep your project rules in AGENTS.md so both agents follow them.
Hands-on test
We haven’t run a hands-on test of this pair yet. This comparison is based on each vendor’s documentation, pricing pages and repositories, linked on the Claude Code and Codex pages. When we add a test, this page will name the task, the tool and model versions, and the date it was run.
