Codex and Claude Code are the same kind of tool at the same $20: an agent that reads your repo, edits files and runs commands. What differs is the meter. OpenAI publishes a rough range for what the plan buys, Anthropic publishes no number at all.
also searched as: codex vs claude code · claude code vs codex · codex vs claude · claude vs codex · codex or claude code · chatgpt codex vs claude code · openai codex vs claude code
Pick Codex if you already pay for ChatGPT, want to try an agent before spending anything, or care that the terminal client is open source and the plan comes with a published estimate.
Pick Claude Code if you already pay for Claude, want one agent that behaves the same in the terminal, the editor, the desktop app and the browser, and would trade allowance for Anthropic's models on a hard change.
Pick neither if you are not going to read the diffs, because both assume someone reviews the code before it merges, and an app builder that hosts the result is the honest tool when you cannot tell whether the change is right.
prices checked 2026-10-05 · by dan scutaru
| Codex | Claude Code | |
|---|---|---|
| price | free tier · from $20/mo | from $20/mo |
| free tier | ChatGPT Free includes Codex for quick tasks: GPT-6 Luna at Standard speed in the desktop app only, subject to rollout. No published message count. | None. Claude's Free plan does not include Claude Code. |
| runs on | web, ios, mac, windows, linux, api | web, ios, android, mac, windows, linux, api |
| top individual plan | ChatGPT Pro at $100, $200 or $500/mo | Claude Max at $100 (5x Pro) or $200 (20x Pro), monthly only |
| published limit at $20 | An estimate, not a cap: 15 to 160 local messages per 5 hours on GPT-6.1 Sol, 350 to 3,000 on GPT-6 Luna | No number. A 5-hour session limit plus a weekly limit. |
| past the limit | Buy ChatGPT credits or use an API key. A running turn may finish. | Wait for the reset, or usage credits at API rates |
| source code | CLI is Apache-2.0. IDE extension and cloud are closed. | Closed, all rights reserved |
At the bottom both cost $20 a month, because Codex comes with ChatGPT Plus and Claude Code with Claude Pro. Codex goes lower, with a light version on Free and the $8 Go plan. Claude Code starts at Pro, or $17 a month paid annually.
Above that the ladders match at $100 and $200, and ChatGPT Pro adds a $500 tier. For teams paying per token, Anthropic's docs put Claude Code at about $13 per developer per active day.
Both meter a five-hour window and a weekly one. OpenAI estimates Plus at 15 to 160 local messages per five hours on GPT-6.1 Sol and says these are not fixed limits. Anthropic gives no figure for Pro, only that Max is 5 or 20 times it.
So the common report that Codex lasts longer at $20 cannot be checked against a published Claude number. What is on record: on 14 September Anthropic ended a summer promotion, which left Claude Code's weekly limit 17% below the promotional level and 25% above where it stood before May.
In the terminal they are closer than the arguments suggest. Both install with one command on macOS, Linux and Windows, read a markdown instructions file, and have skills, hooks, plugins, MCP servers and a scriptable mode for CI.
The differences are at the edges. Codex CLI is open source and accepts an API key or a custom model provider, though an API key loses cloud features such as GitHub code review. Claude Code is closed, built for Claude models, and can bill through Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry.
The threads repeat one trade. People who moved to Codex say it lasts longer at $20, pads less and follows instructions more literally. People who stayed say Claude Code plans better and needs fewer repair passes. One Hacker News commenter had both halves: "Claude's tokens run out much faster. But I get much better results."
The complaints are specific too. About Claude: long comment blocks, and losing the thread when a session compacts. About Codex: plans that do not survive compaction, and fixes more complicated than the problem.
Not across the board, and the answer has flipped more than once this year as each lab shipped a model. Codex is the safer pick if your worry is running out at $20. Claude Code is the pick if you would trade allowance for judgment on a hard change.
Yes, and OpenAI publishes it. The codex-plugin-cc repo adds commands such as /codex:review and /codex:rescue inside Claude Code, so Codex can review Claude's work or take over a task. It needs a ChatGPT account or an API key and counts against your Codex limits.
No. Claude's Free plan does not include it. It starts with Pro at $20 a month, or $17 billed annually. The other route is an API key, where you pay per token. The full answer →
Partly. OpenAI lists Codex on the Free plan with GPT-6 Luna in the desktop app, subject to rollout. The CLI, IDE extension, web and iOS versions, with the stronger models, start with ChatGPT Plus at $20 a month. The full answer →