Is Codex CLI free, and what does it cost?
Codex CLI itself is free and open source, but running it takes a ChatGPT Plus plan or higher ($20 a month, Pro from $100) or an OpenAI API key billed per token. The Free and Go plans include Codex only in the desktop app. On an API key, a median coding-agent day costs about $3.91 on GPT-6.1 Sol, the CLI's default model, and $0.23 on GPT-6 Luna.
What each plan gives you in Codex CLI
| Plan | Price a month | In Codex CLI |
|---|---|---|
| Free | $0 | Not included: Codex runs only in the desktop app, on GPT-6 Luna |
| Go | $8 | Not included: desktop app only, as on Free |
| Plus | $20 | 15 to 160 GPT-6.1 Sol or 350 to 3,000 GPT-6 Luna messages per 5 hours, plus a weekly limit |
| Pro | $100, $200 or $500 | No 5-hour limit on most models; Ultrafast on $500 |
| Business | $20 a seat yearly, $25 monthly | Same as Plus, from 2 seats |
| Enterprise, Edu | Contact sales | Scales with purchased credits |
| OpenAI API key | Per token | No cap: the bill grows with use |
Message ranges are OpenAI's estimates for local tasks. Plus, Pro and Business can buy credits when the allowance runs out. Sources: OpenAI's Codex pricing and the Codex CLI repository, checked October 14, 2026.
What Codex costs on an API key
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.1 Sol, the CLI default | $2 | $0.10 | $10 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
Per million tokens for prompts up to 272K tokens; longer prompts cost $4 and $15 on Sol, $0.20 and $0.75 on Luna. Source: OpenAI API pricing, October 14, 2026.
A month on a ChatGPT plan or an API key
Which one fits you
| If you | Pick |
|---|---|
| Want to try Codex | Free or Go in the desktop app. The CLI starts at Plus |
| Code with Codex a few hours a day | Plus at $20. Set GPT-6 Luna for routine work and the 5-hour window lasts far longer |
| Run out on Plus every week | Pro at $100, which drops the 5-hour limit |
| Run Codex in scripts or CI | An API key, with a spending limit set in your OpenAI account |
| Want a fixed bill and models beyond OpenAI's | A custom provider with a monthly plan: Standard Compute from $19 |
The catch
Every ChatGPT plan price comes with a usage limit. On Plus you need room in both the 5-hour window and the weekly allowance, and the default GPT-6.1 Sol uses the window much faster than GPT-6 Luna. An API key has no limit of its own, so a long agent loop on Sol turns straight into a bill.
Where Standard Compute fits
Codex CLI accepts custom providers that serve the OpenAI Responses API (/v1/responses), and Standard Compute does, at https://api.stdcmpt.com/v1. Run standardcompute launch codex, or add it as a model provider in ~/.codex/config.toml. Plans start at $19 a month for a $20 compute budget, with no overage charges: requests stop at the budget until the plan renews or you upgrade.
FAQ
Is Codex CLI free?
The software is. Install it with npm install -g @openai/codex or Homebrew under the Apache-2.0 license, then sign in with ChatGPT Plus or higher, or paste an OpenAI API key.
Can I use Codex CLI on the ChatGPT Free or Go plan?
No. OpenAI includes Codex on Free and Go only in the desktop app, on GPT-6 Luna at standard speed. The CLI, the IDE extension and Codex on the web start at Plus.
Is ChatGPT Plus enough for Codex CLI?
For a few hours of coding a day, usually. OpenAI estimates 15 to 160 GPT-6.1 Sol messages per five hours on Plus, or 350 to 3,000 on GPT-6 Luna, with a weekly limit on top. If you run out most weeks, Pro at $100 removes the five-hour limit.
How much does Codex CLI cost on an API key?
On our measured coding-agent traffic, a median day costs about $3.91 on GPT-6.1 Sol and $0.23 on GPT-6 Luna at list prices, or about $78 and $5 over 20 working days. A busy day, the 90th percentile, costs about $35.01 on Sol.
What happens when I hit the Codex limit on Plus?
Codex stops until the five-hour window or the week resets. You can buy credits to keep going, upgrade to Pro, or switch to an API key or a custom provider.
Can Codex CLI use other providers or models?
Yes, through a custom model provider in ~/.codex/config.toml, as long as it serves the OpenAI Responses API. That covers OpenAI-compatible gateways such as Standard Compute, which also route to Claude, DeepSeek and Gemini models.