Can you run Codex free in 2026? Yes — and most people never set it up properly.
Codex is OpenAI's coding agent that works like an asynchronous teammate.
You delegate a task, it works independently in a sandbox, and reports back with a diff, logs and citations.
This guide covers exactly what the free tier includes, how to install it, and where the wall is — written from real use, not a spec sheet.
Codex is included in every ChatGPT plan, from Free to Enterprise.
The Free tier sits at $0/month with the smallest allowance, but it is a permanent tier, not a trial.
OpenAI describes it as enough to explore Codex on quick coding tasks, which is an accurate description of its ceiling.
On free you get local coding tasks across four surfaces: the web app, the CLI, the VS Code extension and the iOS app.
What you do not get is the cloud side: no cloud tasks, no automatic GitHub code review, no Slack integration.
Those start at Plus.
The capability underneath is not a toy.
GPT-5-Codex scored 85.5% on SWE-bench autonomous task completion, the open-source CLI has passed 90,000 GitHub stars, and the VS Code extension has crossed 9.8 million installs.
People are not lining up for a toy.
The question is never whether free Codex is capable.
The question is where the free line is drawn.
You do not need an API key, and you should not use one.
An API key switches you to per-token billing the moment you start.
The free path is signing in with your ChatGPT account.
Open chatgpt.com, pick Codex, and sign in with your ChatGPT account.
Pick a repository, write a task, and Codex works in a sandbox and reports back.
Install it with npm install -g @openai/codex.
The CLI is open source under Apache-2.0, so the code is inspectable.
Codex refuses to run outside a git repository, so initialise one first with git init.
Sign in with ChatGPT when prompted, not an API key.
Run a one-shot task with codex exec "your task" and it exits when done.
For building mode use codex exec --sandbox workspace-write "your task", which auto-approves file changes inside the sandbox.
Install the extension, sign in with ChatGPT, and delegate tasks from inside your editor.
This is the most popular surface, with 9.8 million installs behind it.
Start with a small, well-scoped task.
Free-tier limits reward precision, so "fix the date formatting bug in utils/date.ts" goes much further than "refactor the whole app".
⚡ Want an agent workforce that runs without you typing every task? Browse the store → aisuitehq.org/store
The free allowance is capped on a 5-hour rolling window, with weekly limits on top.
OpenAI does not publish one exact free-tier number, but it is the smallest allowance of any plan.
Free users cannot buy credit top-ups.
When you hit the limit, you wait for the window to reset.
That is the practical wall, and it is worth knowing before you start a big refactor.
Cloud features are the second wall.
Automatic GitHub code review, Slack integration and cloud tasks all sit behind Plus and up.
So for an individual developer, free Codex is a genuinely strong deal: a real agent, real delegation, zero spend.
For a team, the free tier is a try-before-you-buy.
The features that turn Codex into a teammate that reviews every pull request while you sleep are paid features.
Codex is an asynchronous coding teammate, closer to handing work to a junior engineer than to autocomplete.
Here is what I actually delegate to it:
My own stack runs Codex inside a wider agent system.
Hermes orchestrates, Codex and Claude Code build, and DeepSeek V4 Flash writes content at $0.14 per million input tokens.
That mix matters, because Codex via API key is not cheap: GPT-5.6 Terra runs at $2 per million input tokens and $12 per million output tokens.
Using the right tool for each layer is how the whole thing stays affordable.
A single orchestrated agent system on one Dell laptop sent 289 emails last week without anyone typing a word.
Codex does the code. The harness decides which model does what.
⚡ Get the full multi-agent setup, done for you → aisuitehq.org/store
The free tier is generous about capability and strict about allowance.
Every one of these mistakes wastes a chunk of your 5-hour window:
I watched a client burn an entire free window in one morning on a single vague prompt.
Scoping the same job into three small tasks finished it inside the same day with allowance left over.
Precision is the free tier's best friend.
And when a task genuinely outgrows the window, that is the signal to buy the plan — not to fight the limit.
Codex is OpenAI-native, runs in your terminal, and is included free with a ChatGPT login.
Claude Code is Anthropic's terminal agent, strongest on logic and technical reasoning, and it pairs well with Codex on the same codebase.
Cursor is an editor-first product: AI inside your IDE, great for interactive coding, but it is a paid subscription, not a free tier with real legs.
The 2026 pattern across the whole category is the same: ship a real free tier so people can prove the value before signing a contract.
For a UK small business or solo operator, the honest verdict is: start with free Codex for the code, and pair it with an orchestrator like Hermes for the emails, content, invoicing and calls.
That combination covers the whole business, not just the repository.
See how the pieces fit in our full Agent OS breakdown.
The 2026 pricing ladder is simple to read:
Students at verified US and Canada universities can claim $100 in ChatGPT credits (2,500 credits) to extend Codex well past the free limits.
One important note: the CLI draws from the same local-message allowance as the web app, so a heavy CLI session and a heavy web session compete for one budget.
And the cheaper GPT-5.4-mini model stretches your message allowance 2.5 to 3.3 times versus the primary models, which matters on the free tier.
Small, precise tasks go further than big vague ones.
Scope one file or one bug per task, and you fit far more work into the same window.
Use GPT-5.4-mini for routine edits so the primary models are saved for hard problems.
Run the CLI inside a clean git repo with a narrow prompt, and review the diff before merging.
Parallelise with git worktrees when you have several independent fixes.
And pair Codex with cheaper engines for everything that is not code.
DeepSeek V4 Flash at $0.14 per million input tokens, Qwen 3.8 Max for long-context writing, and Kimi K2.6 for research keep the monthly bill tiny while Codex handles the engineering.
That is the same layering we deploy for clients — see the OpenAI migration playbook and the n8n workflow library for the non-code half.
My own morning routine is the proof the free path works.
Codex opens the repo, reviews yesterday's diff and fixes the failing test before I finish my coffee.
Meanwhile the marketing agent drafts the day's article and the bookkeeper logs the overnight invoices.
None of that touches the Codex allowance, because the other agents run on DeepSeek V4 Flash and Qwen 3.8 Max.
That is the real trick to running Codex free for a whole business: keep Codex for the code, and let the cheap engines do everything else.
⚡ Stop paying per token. Get a multi-agent system wired to the cheap engines → aisuitehq.org/store
Yes. Codex is included in the ChatGPT Free plan at $0/month, covering local coding tasks on the web app, CLI, VS Code extension and iOS app, with the smallest usage allowance of any plan.
The free allowance is capped on a 5-hour rolling window with weekly limits on top. It is enough for quick, well-scoped coding tasks, and free users cannot buy credit top-ups.
No. Sign in with your ChatGPT account and you stay on the free allowance. An API key switches you to pay-per-token billing with no window, but you pay for every token.
Yes. The Codex CLI is open source under Apache-2.0 and works on the Free plan when you sign in with ChatGPT. Install it with npm install -g @openai/codex and run it inside a git repository.
Plus at $20/month unlocks cloud tasks, automatic GitHub code review, Slack integration and credit top-ups when you hit a limit. Free covers local coding tasks only.
Start free and prove the value on real work first. Upgrade to Plus or Pro the day Codex is doing daily work for you, not a day before.
Codex is one layer.
A full agent workforce is the OS underneath it.
We build both for UK businesses, from £149 setup with retainers from £199/month.
🚀 See the store → aisuitehq.org/store