Can you use Codex for free? What you can and can't do on the free plan, and the actual cost of paid plans and API keys

Yes, you can use Codex for free because Codex is included in ChatGPT's free plan (Free). However, OpenAI's pricing page lists just one item for the free plan: access to GPT-6 Luna at standard speed in the desktop app (rolling out gradually). Go at $8 per month offers the same, while Codex for CLI, IDE extensions, and the web is listed starting with Plus at $20 per month. To run Codex CLI without a plan, you can pay for the tokens you use with an API key. This is the opposite of Claude Code, which is unavailable on the free plan.

This page organizes what you can and cannot do on the free plan according to OpenAI's documentation, and compares the free plan, Plus, API keys, Kunavo, and Claude Code in a single table. It also shows the actual cost of a task when paying with an API key. For example, GPT-6 Luna, the same model as in the free plan, costs $0.04 per 1M input tokens and $0.20 per 1M output tokens on Kunavo, or about $0.025 for one 20-step task.

What you can and cannot do on the free plan

The following summarizes the information on OpenAI's Codex pricing page (English, checked on October 10, 2026). There is only one line under the free plan, so comparing it with the other plans makes its scope clear.

For reference, Plus is listed with an estimate of 350–3,000 local messages with GPT-6 Luna per 5 hours. OpenAI describes this as an estimate, not a fixed limit, and you can check your account's limit and reset time on the Codex usage page. The pricing page also describes the intended use of each plan in one line: the free plan is “try Codex features on short coding tasks,” Go is “for light coding,” and Plus is “for a few focused coding sessions per week.” OpenAI's own description positions the free plan as an entry point for trying Codex.

Codex pricing plans

Prices are in US dollars as listed on the pricing page (October 10, 2026). Paid plan usage is divided into time windows. Only Plus and Standard Business have published 5-hour estimates; Pro currently has no 5-hour limit, and weekly limits may apply to any plan. After you use the included allowance, Plus and Pro let you continue working with credits. See Codex pricing for the break-even point between a plan and pay-as-you-go usage, and how to continue after using up the allowance.

Comparing the free plan, Plus, API keys, Kunavo, and Claude Code in one table

Here, “one task” means 20 steps of agent work, with each step calculated as 25,000 input tokens and 1,200 output tokens. This is the common unit used on this site's Codex and Claude Code pricing pages, allowing comparisons across the pages. Plans do not have a per-task price; instead, they include usage allowances.

This table shows two things. First, GPT-6 Luna, the model on the free plan, is inexpensive even when purchased on a pay-as-you-go basis: about $0.025 per task on Kunavo, and the minimum $10 top-up covers about 400 tasks. Second, the free plan's main limitation is where you can use Codex, rather than the model. The free plan listing on the pricing page only includes the desktop app, so if you want to use Codex in a terminal or editor, you need to choose between subscribing to Plus and paying with an API key. Information about Claude Code is based on Claude's pricing page and Anthropic's Help Center article (both checked on October 10, 2026).

How much does one task cost when paying with an API key?

API key charges are based on tokens used multiplied by the unit price, with no monthly fee. Kunavo's prices are read from the catalog each time they are displayed; OpenAI's prices are from OpenAI's API pricing page (October 10, 2026).

If you run 5 tasks per day for 20 days, or 100 tasks per month, at Kunavo's rates, GPT-6 Luna costs about $2.48, GPT-6 Sol about $49.60, and GPT-5.6 Sol about $128.80. There is no charge in months when you do not use it. Actual costs may be higher or lower. Prompt caching can reduce costs: for example, cached input for GPT-6 Sol is billed at $0.08 per 1M tokens instead of $0.80. Conversely, reasoning tokens for reasoning models are billed as output, increasing costs, and provider-added context may be counted as input on Kunavo's GPT route. Check the usage history in your dashboard for the final amount.

Run Codex CLI on pay-as-you-go usage — Kunavo setup

Kunavo is an independent API gateway, separate from OpenAI. Codex CLI only supports custom providers through the Responses API, but Kunavo provides POST /v1/responses as is, so it works simply by adding one provider to the configuration file. Create an account, top up your balance, and create a key on the API keys page, then write the following block to ~/.codex/config.toml. This is the exact configuration shown in the Codex CLI integration documentation (English).

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

The value of env_key is the name of the environment variable that holds the key, not the key itself. The key is passed through the environment variable, so it is not stored in the configuration file. Then set the key using that variable name and launch the CLI. Add -m to switch models for that launch only, without changing the configuration file.

# 変数名は config.toml の env_key と同じにする
export KUNAVO_API_KEY=sk-kn-...

codex                     # 既定のモデル（gpt-5-6-sol）で起動
codex -m gpt-6-luna       # 無料プランと同じ GPT-6 Luna を従量課金で
codex -m gpt-6-sol        # 重めの作業は GPT-6 Sol で

Your Kunavo balance covers only usage through this API. It does not add to your ChatGPT plan's allowance or reset its limits, and it does not enable the Codex web app or cloud features. Kunavo operates with shared capacity and offers no dedicated capacity or SLA. For use cases that require guarantees, contract directly with OpenAI. See How to use Codex for instructions from installation through your first task.

Paying from Japan

You can top up your Kunavo balance starting at $10 using cards, including JCB, Apple Pay, or Google Pay. There is no monthly fee, and balances do not expire. PayPay and convenience store payments are not supported. Note that this payment is for API usage, not a ChatGPT plan. Subscribe to plans such as Plus directly through OpenAI. See Claude payment methods for details on Kunavo payment methods, receipts, and invoices (the balance is shared across Claude and GPT models).

How Claude Code differs — whether it can be used for free is the reverse

Codex’s cheapest entry point is a free plan, but Claude Code has no free plan. The Claude pricing page does not include Claude Code in the free plan; it is available with the Pro plan ($20/month, or $17/month when billed annually) or higher, or with an API key. Beyond that, the setup is similar to ChatGPT: Pro and Max have five-hour session limits and weekly limits. If you pay with an API key, one 20-step task using Claude Sonnet 5 costs about $0.87 with Kunavo and about $1.24 at Anthropic’s list price.

The only way to try Claude Code for free is with a guest pass. Eligible Pro or Max subscribers can share this seven-day Pro trial, which includes Claude Code, but it is available only to people who have never had a paid Claude plan. You must provide payment information when signing up; unless you cancel, the trial automatically converts to a paid Pro plan. See Claude Code pricing for the full pricing details and Codex vs. Claude Code for a comparison of the two tools.

Frequently asked questions

We checked the official OpenAI and Anthropic pages October 10, 2026. Check your account page for the latest usage limits and reset times.