Is Codex free? Codex pricing, plans and usage, and what a session costs

Codex is free to start: the Codex CLI is open source, and ChatGPT's Free and Go plans include Codex in the desktop app, subject to rollout. The cheapest plan that lists Codex in the CLI, the IDE extension and the web is ChatGPT Plus at $20 a month; an API key, billed per token, works too. To check your Codex usage, open chatgpt.com/codex/settings/usage or run /status in the CLI.

Is Codex free?

Partly. On the Codex pricing page, read on 18 October 2026, the Free plan says "Explore Codex capabilities on quick coding tasks", with GPT-6 Luna at standard speed in the desktop app, subject to rollout. Go, at $8 a month, gives the same model in the same app "for lightweight coding tasks".

The Codex CLI itself is free software (Apache-2.0), but the model behind it is not. OpenAI's pricing page lists the CLI from Plus up; on Free and Go it names only the desktop app. On an API key there is no plan at all: every token is billed at OpenAI's API rates, and any user can run extra local chats that way.

Codex pricing: which ChatGPT plans include it

Codex has no price of its own: it comes with a ChatGPT plan, or runs on an API key. The plan table at the top of this page is OpenAI's Codex pricing as read on 18 October 2026.

• Plus, $20 a month: Codex on the web, in the CLI, the IDE extension and iOS, with cloud integrations such as automatic code review and Slack.
• Pro, $100, $200 or $500 a month: currently no five-hour limit; the $500 tier adds Ultrafast mode.
• Business, $25 a user a month, or $20 billed annually: two or more users, ChatGPT and Codex in the desktop and mobile apps, larger cloud VMs and credits.
• Enterprise and Edu: priced by OpenAI's sales team.

At the limit, OpenAI lets a turn already running finish. Plus and Pro can then buy credits to keep working, and anyone can switch the CLI to an API key.

Codex API pricing, per token

With an API key, OpenAI "bills API key usage through your OpenAI Platform account at standard API rates" — the per-token table at the top of this page, read on 18 October 2026. GPT-6.1 Sol, for example, is $2 per million input tokens and $10 output. Above 272K input tokens the long-context rate applies: double for input, 1.5× for output. How those rates compare with Claude's is in Codex vs Claude Code.

How to check your Codex usage

Open the usage dashboard at chatgpt.com/codex/settings/usage for your account's five-hour and weekly windows and their reset times, and run /status in the terminal for the Codex CLI session you are in.

The dashboard shows your current limits and reset times for the whole account. There are two windows:

• A five-hour period. Plus and Standard Business plans get an estimated number of local messages per five hours, which depends on the model; Pro plans currently have no five-hour limit. Local messages and cloud chats share the allowance.
• A weekly limit. "Weekly limits may also apply" — the page gives no figure for it.

OpenAI calls the per-plan figures estimates and changes them, so read them on the Codex pricing page and not from a copy here. Both windows by plan, what to do at the limit, and what one machine's windows recorded: Codex usage limits. The dashboard is one account's. What an admin sees of a whole workspace on ChatGPT Business and Enterprise, and what no console shows: Codex analytics for a team.

Check Codex usage from the terminal

Two slash commands cover it:

• /status — "Display session configuration and token usage": the model, approval policy, writable roots and remaining context. The pricing page points to it for "your remaining limits during an active Codex CLI session".
• /usage — "View account token usage or use a rate-limit reset". Add daily, weekly or cumulative to pick the period.

/statusline puts rate limits, context and token counters in the footer so you can see them all the time. To get back to an older session, codex resume opens a picker; --last jumps to the most recent one in the current directory, and --all drops that filter.

Every one of these is about your plan, or the session you are in. None of them shows tokens or cost per project, or Codex next to the other agents you run.

What Codex costs, session by session

On a ChatGPT plan, Codex costs what the plan costs. OpenAI's own words: "API token prices are separate from subscription usage; don't use them to estimate included tasks." So any dollar figure per session is an equivalent, not a charge.

On my machine, read on 3 October 2026, the median Codex call came to $0.032 at list prices across 16,415 calls, against $0.094 for a Claude Code call. Cache reads were 73% of Codex's cost: every call re-reads the conversation so far.

Caprock gives you that equivalent. It reads Codex's session files from $CODEX_HOME (~/.codex by default) — sessions/ and archived_sessions/ both, compressed .jsonl.zst files included — from the CLI and the desktop app alike, with no hook and nothing written into Codex's config. Each turn is priced from Caprock's own table of OpenAI's list prices, and Codex sits next to Claude Code on the same screens — Codex vs Claude Code reads the two side by side:

⤢

• Model mix — every GPT model's turns, tokens and cost beside Claude's. The screen says "At API list prices", and on a flat plan adds that this is not money out of pocket.
• Per project — Codex and Claude Code work on the same repository add up in one row.
• Background usage — Codex's automatic approval reviewer is counted in tokens on its own line, with no price: OpenAI publishes none for it.
• Plan limits — Codex's 5-hour and weekly windows sit under a Codex heading in the Plan limits panel, beside Claude Code's, with the time Codex wrote them. Only the windows your plan has get a row, and nothing is forecast from them.
• Memory — what Codex wrote back is searchable beside Claude Code's replies.
• Export — caprock export --agent codex writes the Codex rows out as TSV, CSV or JSON lines.

Caprock reads Codex sessions from their files whether or not it started them, and types only into the ones it started. Every dollar is a list-price estimate, as Claude Code's own /usage is, beside Codex's own windows. Free, open source; nothing leaves your machine. Install Caprock to see them alongside your other agents.

Where the numbers stop

• The limits are as old as the last Codex session. Codex writes them into its transcripts, not to a live feed, so with no session running the figure can be hours old. The "as of" time says so.
• Some turns report only a total. Codex sometimes writes a token total with no split into input, cached and output. Caprock prices that total as input, so for those turns the cost is an upper bound.
• Fast mode is priced at the standard rate. Recent Codex versions record the service tier in the transcript, but Caprock does not read it yet, so a fast-mode turn costs the standard price.

Which tools show Codex CLI usage and cost?

The usage dashboard and /status are OpenAI's own and show your limits; ccusage codex prints cost tables in the terminal, with Codex marked Beta; and Caprock, a free desktop app, shows Codex's cost per session beside Claude Code.

• ccusage reads Codex as a data source marked Beta: ccusage codex daily, monthly and session print tables in the terminal, and its plain ccusage daily includes Codex too. It reads CODEX_HOME with archived and compressed sessions, prices from LiteLLM's dataset as "an API-equivalent estimate", and has a --speed option for fast mode. Its Codex page says nothing about plan limits. If you want a cost table when you ask for one, with nothing running in between, use it.
• The usage dashboard and /status are the official word on your limits. If that is the only question, you need nothing else.
• Caprock is a free desktop app: Codex beside Claude Code and the other agents you run, whichever terminal started them, cost per repository, and Codex's limits next to Claude Code's. Use it when you want every agent's limits and spend in one record — five agents in one table shows what that looks like.