Is Codex free? Yes, with limits — the free tier, the plans, and what a task costs

Yes, Codex has a free tier, and that is exactly where it differs from Claude Code. OpenAI's Codex pricing page lists ChatGPT Free at $0 a month with a single line for Codex: “GPT-6 Luna at Standard speed in the desktop app, subject to rollout.” The $8 Go plan carries the same line. The CLI, the IDE extension and Codex on the web are listed from Plus, at $20 a month; the CLI and the IDE extension also run without any plan, on an API key billed per token. Claude Code has no free tier: Claude's Free plan does not include it.

So the useful questions are how far the free tier reaches and what each step past it costs. This page answers both from OpenAI's own pricing page, puts every route in one table with what a real coding task costs on it, and ends with the per-token route for Codex CLI. On that route GPT-6 Luna, the Free plan's model, is $0.04 input and $0.20 output per million tokens on Kunavo: about $0.025 for a 20-step task.

What the free tier includes, and what it leaves out

Everything in this section comes from OpenAI's Codex pricing page, read on October 10, 2026. The Free plan's card has one line, so the edges of the free tier show up in what the page says about the other plans.

• Model: GPT-6 Luna at Standard speed. No other model is listed for Free or Go.
• Where it runs: the desktop app. The CLI, the IDE extension, Codex on the web and iOS appear on the Plus card, not on Free or Go.
• Who has it: “subject to rollout” means it reaches accounts in stages, so yours may not have it yet.
• How much: no allowance is published for Free or Go. OpenAI's five-hour estimates cover Plus and Standard Business only (on Plus, 350–3,000 local GPT-6 Luna messages per five hours, which it calls estimates rather than fixed limits), and weekly limits may also apply. The Codex usage dashboard shows your own account's limits and reset times.
• Not included: image generation, which “isn't available on the Free plan”.
• Past the limit: buying credits is described for Plus and Pro, not for Free. What the page says of all users is that they “may also run extra local chats using an API key”, billed at standard API rates.

OpenAI's one-line pitch for each tier says the same in its own words: Free is to “explore Codex capabilities on quick coding tasks”, Go to “use Codex for lightweight coding tasks”, and Plus to “power a few focused coding sessions each week”. The free tier is for trying Codex; regular work is what the plans and the keys are priced for.

Codex plans and prices

US dollars, as listed on October 10, 2026. On the plans, usage runs in windows: OpenAI's estimates show “local messages per five-hour period for Plus and Standard Business”, “Pro plans currently have no five-hour limit”, and weekly limits may apply on top. Once the included limits are used, credits keep Plus and Pro working. When Codex stops mid-task, Codex usage limits works out which limit it hit and what fixes it.

Every route side by side, with Claude Code for comparison

One coding task here is 20 agentic steps of about 25,000 input and 1,200 output tokens each, the unit the other Codex and Claude Code cost pages on this site use, so the figures compare across them. A plan has no per-task price; it has a usage window instead.

Two things stand out. The model the free tier runs is cheap to buy outright as well: at Kunavo's rate a 20-step GPT-6 Luna task is $0.025, so the $10 minimum top-up covers roughly 400 of them. And what the free tier lacks is reach more than model, because its card lists the desktop app and nothing else. If you want Codex in a terminal or an editor, the choice is between a plan and a key, and Codex pricing works out the break-even between them. The Claude Code row comes from claude.com/pricing and Anthropic's help article on Claude Code with Pro or Max, both read on October 10, 2026.

What a Codex task costs on an API key

On a key the bill is tokens times rate, with no monthly fee. Kunavo's rates are read from its catalog when this page renders; OpenAI's come from its API pricing page, read on October 10, 2026.

A heavy month, five such tasks a day for 20 days, comes to $2.48 on GPT-6 Luna, $49.60 on GPT-6 Sol and $128.80 on GPT-5.6 Sol at Kunavo's rates, and a month you do not code costs nothing. Those figures move in both directions. Prompt caching lowers them: cached input on GPT-6 Sol bills at $0.08 per million instead of $0.80. Reasoning tokens raise them, because they bill as output, and Kunavo's GPT routes can report context the provider adds as input. Read the billed usage in your dashboard before trusting any estimate, this one included.

Running Codex CLI per token through Kunavo

Codex CLI accepts a custom provider only over the Responses API (wire_api has one legal value, "responses"), and Kunavo, an independent gateway rather than OpenAI, serves POST /v1/responses natively. Create an account, top up, make a key in the dashboard, and put this block, the one on Kunavo's Codex CLI setup page, in your Codex config:

model = "gpt-5-6-sol"
model_provider = "kunavo"

[model_providers.kunavo]
name = "Kunavo"
base_url = "https://api.kunavo.com/v1"
env_key = "KUNAVO_API_KEY"     # the NAME of the variable, not the key
# wire_api defaults to "responses", which is the only supported value

env_key holds the name of an environment variable, not the key, so the file is safe to commit. Export the key under that name and start Codex; -m picks the model for one session without touching the file:

export KUNAVO_API_KEY=sk-kn-...   # the variable env_key names

codex                     # GPT-5.6 Sol, the block's default
codex -m gpt-6-luna       # GPT-6 Luna, the Free plan's model, per token
codex -m gpt-6-sol        # GPT-6 Sol, for heavier tasks

There is no monthly fee, top-ups start at $10, and the balance does not expire. Know what it does not buy: a Kunavo balance pays for these API calls only. It adds nothing to a ChatGPT plan's allowance, does not reset a plan's limit, and does not unlock Codex on the web or the cloud features. Capacity is shared, with no dedicated quota and no SLA; if you need guarantees, buy from OpenAI directly. Getting the CLI onto your machine first is covered in how to install Codex CLI.

Codex and Claude Code answer this question in opposite ways

Codex's cheapest door is a free plan. Claude Code has none: Claude's pricing page shows no Claude Code on Free, and access starts with Pro at $20 a month ($17 billed annually) or with an API key. Past that, Claude's plans meter usage much as ChatGPT's do, since Pro and Max each have “a five-hour session limit and a weekly limit”. On a key, a 20-step Claude Sonnet 5 task is $0.87 on Kunavo against $1.24 at Anthropic's list price.

The one way to try Claude Code without paying is a guest pass: an eligible Pro or Max subscriber can send a seven-day Pro trial that includes Claude Code to someone new to Claude's paid plans. It asks for payment details at signup and becomes a paid Pro plan unless cancelled. The Claude side in full is on Is Claude Code free?

Frequently asked questions

OpenAI's and Anthropic's pages checked on October 10, 2026. Your own account's dashboard is the authority on current limits and reset times.