Short answer: choose Claude Code for hands-on, multi-step work in your terminal with Claude models (it leads Terminal-Bench 4.0 at 64.8%); choose OpenAI Codex if you already have ChatGPT and want to delegate tasks to cloud sandboxes that return PRs, plus built-in GitHub code review. Both start at $20/month, and Codex has limited free use.
Side by side
| Feature | Claude Code | OpenAI Codex |
|---|---|---|
| Made by | Anthropic | OpenAI |
| Status | Active | Active |
| Type | CLI / terminal, IDE extension, Cloud agent | CLI / terminal, Cloud agent, IDE extension, Code review |
| Interfaces | Terminal · VS Code / JetBrains · Desktop app · Web | Terminal (Codex CLI) · IDE extension · Web / ChatGPT · Desktop app · iOS |
| Open source | No | ✓ Codex CLI is open source (Apache-2.0); cloud service proprietary |
| Free tier | — No | ✓ Yes |
| Starting price | from $20/mo | Free tier · from $20/mo |
| Plans | Pro: $20/mo ($17/mo billed annually) Max: From $100/mo Team: Per seat Enterprise: $20/seat/mo + usage at API rates API (pay-as-you-go): Per token | ChatGPT Free / Go: Included (limited) Plus: $20/mo Pro: $100, $200 or $500/mo Business / Enterprise: Per seat API key: Pay per token |
| Models | Anthropic Claude models (Opus, Sonnet, Haiku families) | OpenAI GPT models (current GPT-6 family in Codex) |
| Bring your own key | ✓ Yes | ✓ Yes |
| Agent mode | ✓ Yes | ✓ Yes |
| Autocomplete | — No | — No |
| Cloud / background agents | ✓ Yes | ✓ Yes |
| PR code review | ✓ Yes | ✓ Yes |
| MCP support | ✓ Yes | ✓ Yes |
| Self-host option | — No | — No |
| Best for | Developers who are comfortable in the terminal and want an agent that can own multi-step tasks end to end. | ChatGPT subscribers who want to delegate well-scoped tasks to background agents and review PRs. |
| Last verified | 2026-10-10 | 2026-10-10 |
The real difference: local loop vs delegation
Claude Code is built around an interactive local loop. It explores your repo, plans, edits files and runs commands in your terminal (also IDE, desktop and web surfaces), with MCP servers, hooks and subagents to extend it.
Codex is built around delegation. The CLI and IDE extension work locally, but its distinctive mode is the cloud: each task runs in its own sandbox in parallel and comes back as a diff or pull request, with automatic GitHub code review and Slack integration.
Where Claude Code wins
- Benchmarks — 64.8% vs 58.2% on Terminal-Bench 4.0 (Oct 2026, max effort).
- Deep interactive sessions — strong at multi-file changes in large, unfamiliar codebases.
- Extensibility — MCP, custom slash commands, hooks and subagents.
Where Codex wins
- Free entry — included (limited) in ChatGPT Free and Go; Plus is $20/month.
- Parallel cloud tasks — fire off several tickets and review PRs later.
- Open-source CLI — Apache-2.0, can run with a plain OpenAI API key.
Pricing compared
Claude Code: Claude Pro $20/month ($17/month annually), Max from $100/month, Team per seat, Enterprise $20/seat + API usage, or pay-as-you-go API. Codex: included in ChatGPT Free/Go (limited), Plus $20/month, Pro $100, $200 or $500/month, Business/Enterprise per seat, or an API key (CLI/IDE only); usage beyond the plan allowance is billed in credits. Verified 2026-10-11 on the official pricing pages.