How to Use GPT-5.5 for Free with Codex?

OpenAI shipped GPT-5.5 on April 23, 2026 and, as part of the launch, rolled Codex out to every ChatGPT plan including Free and Go for a limited time. That one line in the announcement is the fastest free path to the new model: install the Codex CLI, sign in with a ChatGPT account, and drive GPT-5.5 from the terminal with no API key and no credit card.

This guide walks through the install, the authentication options, the model switch, the caps you will hit, and how to wire Codex into a real coding workflow. For the underlying model overview, see What is GPT-5.5. For other free paths (trial credits, aggregators), see our GPT-5.5 for free guide.

TL;DR

• Codex CLI runs GPT-5.5 against your local repo with a 400 K context window.
• ChatGPT Free, Go, Plus, Pro, Business, Enterprise, and Edu all get Codex access; Free and Go are limited-time.
• Install with npm install -g @openai/codex or brew install codex.
• Sign in via ChatGPT OAuth in a browser, or use device-code flow on headless boxes.
• Switch models mid-session with /model gpt-5.5; check quota with /status.
• Pair the CLI with Apidog so the API calls you will eventually run in production are pre-built and tested.

Why Codex is the easiest free path

OpenAI’s API has a paid-only default; GPT-5.5 on the Responses endpoint costs $5 per million input tokens and $30 per million output tokens once it goes general. Codex short-circuits that by wrapping the same model in a CLI that authenticates against a ChatGPT account instead of an API key. The plan tier determines the rate limit; the model running underneath is the real GPT-5.5.

Install Codex CLI

Two supported install paths.

# npm (cross-platform)
npm install -g @openai/codex

# or Homebrew (macOS / Linux)
brew install codex

Verify the install:

codex --version

You should see a version at or above 0.28.0; older releases do not list GPT-5.5 in the model picker.

Authenticate with a ChatGPT account

Run the CLI for the first time and it prompts for a sign-in method.

Browser OAuth (local machines)

codex

A browser tab opens. Sign in with the same email you use for ChatGPT. The CLI caches the session so subsequent runs skip the prompt.

Device code flow (headless boxes, remote servers)

codex login --device-auth

The terminal shows a short code and a URL. Open the URL on any other device, paste the code, confirm. The headless machine finishes sign-in when the confirmation lands.

API key fallback

If you already have a paid API key and prefer it:

printenv OPENAI_API_KEY | codex login --with-api-key

This path charges the key’s billing account instead of the ChatGPT plan. Use it if you want Codex but want the usage to land on a team’s paid API org.

Pick GPT-5.5 as the model

Codex defaults to whatever OpenAI marks as “recommended” for your plan. On all paid plans that is already gpt-5.5. On Free and Go you may need to switch manually.

Mid-session

/model gpt-5.5

The CLI prints the active model and the plan-specific rate limits.

From the launch flag

codex --model gpt-5.5

Check remaining quota

/status

The output shows weekly message budgets, context window, and the expiration time of the current Free/Go trial window if you are on one.

First session: a realistic example

Codex launches into a full-screen terminal UI that can read your repo, run commands, and edit files. A normal first session looks like this:

cd ~/Projects/my-app
codex --model gpt-5.5

Inside the session:

> Read README.md, then open scripts/deploy.sh and summarize what it does in five bullets.

Codex opens the file, summarizes, and prints the bullets. Follow-up:

> Refactor deploy.sh so it exits on any failed step, and add a dry-run flag. Keep backwards compatibility.

GPT-5.5 proposes a diff, waits for approval, and applies it. Run tests inline:

> Run the deploy test suite and show me the failing case.

The CLI streams the test output back into the session. If anything fails, ask the model to fix it; the loop continues until green.

This is the workload GPT-5.5 is tuned for. The OpenAI launch post puts the SWE-bench score at 88.7 %, up from 74 % on GPT-5.4; the jump shows up in exactly this multi-file, tool-using, terminal-driven work.

What Codex gives you that raw API calls do not

Four things the CLI wraps around the model:

• Repository context. Codex reads the file tree, indexes relevant files, and gives the model file handles rather than pasted content. The 400 K context window stays mostly empty for tool calls and output.
• Command execution with approvals. The model suggests a command; you approve before it runs. This is the safety control that keeps agentic coding from flooding your shell.
• Diff previews before writes. Every file edit previews as a unified diff; reject, accept, or edit before it commits.
• Session persistence. Each project gets a session history. Reopen the CLI tomorrow and the context picks up where you left off.

Without Codex you would rebuild each of these yourself on top of the Responses API. Our GPT-5.5 API guide shows the bare-metal version if you want to, but for day-to-day coding the CLI wins on ergonomics.

Rate limits and caps per plan

Exact numbers rotate, but the shape on April 23, 2026 looks like this.

When you hit the cap, Codex returns a plan-specific error instead of silently degrading. /status shows the exact remaining count.

Editor and IDE integration

The same Codex login backs the VS Code extension, the JetBrains plugin, and the Codex cloud app. After signing in via the CLI, the IDE extension reuses the cached credentials; no second sign-in.

For Apidog users, the useful pairing is:

1. Prototype a request inside the Codex CLI (run the GPT-5.5 prompt against this file).
2. Export the prompt and structured output into an Apidog collection for sharing with teammates.
3. Swap the Codex path for a direct API call once the contract stabilizes and you move to a paid key.

See Apidog in VS Code for how to wire the collection into the same editor you are running Codex inside.

Keeping the workflow safe on Free and Go

Two guardrails worth setting on Day 1:

• Always-approve file writes. In ~/.codex/config.json, set "autoApproveWrites": false. The default is already safe on Free, but on some Go-tier setups Codex will auto-apply diffs it considers trivial.
• Scope the workspace. Run codex from the project directory only. The CLI’s read scope starts from the working directory; opening it at ~ gives it the entire home folder.

OpenAI ran GPT-5.5 through third-party safety review before launch, and the CNBC launch coverage spells out the cyber-specific red-teaming. Even so, the CLI is still running on your machine; the usual rules about reviewing diffs before applying them still apply.

When to move off the free path

The “limited time” framing in OpenAI’s announcement means Codex Free and Go will eventually require an upgrade. Three triggers to plan for:

1. Quota exceeded weekly. You are doing real work; time to move to Plus or Pro.
2. You want the direct API. Your workload has grown past the CLI; the GPT-5.5 API guide covers the migration.
3. Your team needs seat-level billing. Business or Enterprise via the pricing breakdown.

Each migration keeps the underlying model the same; only the billing and surface change.

FAQ

Does Codex run GPT-5.5 Pro too?Pro is not currently exposed inside Codex. The CLI uses the standard GPT-5.5 model across every plan; Pro stays on ChatGPT and (eventually) the direct API.

Can I use Codex without a ChatGPT account?No. Either a ChatGPT sign-in or an OpenAI API key; Codex needs one of them. The free path depends on the ChatGPT sign-in.

How long will Free and Go access last?OpenAI explicitly called it “for a limited time.” Assume weeks to a few months, and plan to upgrade if the usage grows.

Does Codex work offline?No. Every GPT-5.5 call is a round-trip to OpenAI’s infrastructure.

How is this different from the ChatGPT web app?Codex runs inside a terminal session with your local filesystem, shell access, and repo context. The web app does not have any of those.