OpenAI Codex April 2026 Update: GPT-5.5, Browser Use, and CLI 0.124–0.125

OpenAI shipped a wave of Codex updates in April 2026, headlined by GPT-5.5 — the newest frontier model — landing in Codex as the recommended default for most coding and knowledge-work tasks. The Codex app added in-app browser use for local development servers, automatic approval reviews routed through a reviewer agent, and the Codex CLI moved through 0.124.0 and 0.125.0 with quick reasoning controls and reasoning-token reporting in codex exec --json.

TL;DR

• GPT-5.5 is OpenAI's newest frontier model and is now selectable in Codex for complex coding, computer use, knowledge work, and research workflows.
• OpenAI recommends GPT-5.5 for most Codex tasks when visible in the model picker — especially implementation, refactors, debugging, testing, validation, and knowledge-work artifacts.
• Switch with codex --model gpt-5.5 or /model in the CLI; use the model selector in the IDE extension and Codex app.
• Browser use in the Codex app drives local dev servers and file-backed pages for visual bug reproduction and fix verification, managed via a bundled Browser plugin with allowed/blocked website lists.
• Automatic approval reviews route eligible approvals through a reviewer agent before execution.
• Codex CLI 0.125.0: codex exec --json reports reasoning-token usage.
• Codex CLI 0.124.0: quick reasoning controls; model upgrades reset reasoning to the model default; eligible ChatGPT plans default to the Fast service tier unless opted out.

Quick Answer

If you use OpenAI Codex, open the model picker and switch to GPT-5.5 for your next implementation, refactor, or debugging task. From the CLI, run codex --model gpt-5.5 or use /model mid-session. If you work against a local dev server, enable browser use in the Codex app to let the agent reproduce visual bugs and verify fixes against localhost.

What Shipped in April 2026

GPT-5.5: When and How to Use It

GPT-5.5 is OpenAI's new frontier model. In Codex, OpenAI calls it out as the right choice for the bulk of day-to-day coding agent work and longer-horizon knowledge tasks:

• Implementation of new features across multiple files
• Refactors that need to hold a larger plan in mind
• Debugging — including the kind that requires reading test failures and reasoning across modules
• Testing and validation workflows
• Knowledge-work artifacts like specs, design docs, and migration plans

Switching is the same wherever you run Codex:

• CLI: codex --model gpt-5.5 to launch a session on the model, or /model inside a running session to switch
• IDE extension: model selector
• Codex app: model picker

If GPT-5.5 is not yet visible in your picker, it is rolling out — keep your CLI and extension current.

Browser Use in the Codex App

The Codex app gained the ability to drive an in-app browser against your local development server and file-backed pages. Practical uses:

• Visual bug reproduction — describe a UI bug and let Codex open the page and see what you see.
• Verifying local fixes — after Codex edits a component, it can reload the dev server and confirm the change.
• Smoke-testing flows that depend on a real DOM or rendered output, not just unit tests.

Browser use is managed through the bundled Browser plugin with allowed and blocked websites so you control where the agent can navigate. For agent workflows that already touch parallel sandboxes, this closes the loop between code edits and what the user actually sees.

Automatic Approval Reviews

Where configured, Codex now routes eligible approval prompts through an automatic reviewer agent before they reach you. The reviewer evaluates the proposed action and surfaces its assessment alongside the approval, so high-confidence operations can clear faster while risky ones still escalate.

Codex CLI: 0.124.0 and 0.125.0

The CLI moved fast in late April:

• 0.125.0 — codex exec --json now reports reasoning-token usage, which makes it easier to budget runs and tune reasoning effort in scripts and CI.
• 0.124.0 — adds quick reasoning controls for changing reasoning effort mid-session; model upgrades reset reasoning to the model default so you don't carry a stale setting across versions; and eligible ChatGPT plans default to the Fast service tier unless you opt out.

If you script Codex from CI, the --json reasoning-token field is the upgrade you actually want — it makes cost and latency observable per run.

How This Compares to Other Agents

For deeper background see our OpenAI Codex desktop app guide, the AI coding agents explainer, and the OpenAI Codex tool page.

Sources

• OpenAI Codex changelog (official): developers.openai.com/codex/changelog
• OpenAI Codex product page: developers.openai.com/codex
• OpenAI Codex CLI: github.com/openai/codex

For broader model context, see our Claude Opus 4.6 vs GPT-5.1 Codex Max vs Gemini 3 Pro comparison and the beginner's guide to AI models.