OpenAI Codex on Mobile: Coding Agents in Your Pocket

TL;DR

On May 14, 2026, OpenAI moved Codex into the ChatGPT mobile app on iOS and Android — in preview, across every plan from Free to Enterprise¹. The phone pairs with a Codex desktop session via QR code, and at launch it only connects to the macOS host app; Windows support is on the way². The same release made Remote SSH and Hooks generally available on all tiers, and shipped programmatic access tokens for Business and Enterprise customers. Codex now serves more than 4 million weekly users — and at its 2-million-user milestone in March 2026, the platform was already 5x larger than three months earlier and growing more than 70% month over month³. With Codex mobile in preview, Anthropic's Claude Code Remote Control already on iOS and Android (Pro and Max plans), and Cursor's cloud agents drivable from the web, the "always-on coding agent" pattern is now table stakes.

What You'll Learn

• What the May 14, 2026 Codex mobile release actually does — and what it does not do
• How the mobile preview interacts with Remote SSH, Hooks, and access tokens
• How Codex on mobile compares with Claude Code Remote Control and Cursor cloud agents
• What Codex's 2M-to-4M weekly-user growth from March through April suggests about coding-agent adoption
• The practical limits around regions, plan tiers, and HIPAA workspaces

What Shipped on May 14

OpenAI's announcement page is titled "Work with Codex from anywhere," and the framing is accurate: nothing about the local Codex experience changed, but the surfaces from which you can drive it expanded. The mobile rollout started Thursday, May 14, 2026, on iOS and Android, in preview, across all ChatGPT plans including Free and Go, in every supported region¹.

Pairing is QR-based: open Codex on your Mac, select "Set up Codex mobile" in the sidebar, scan the QR code with your phone, and confirm the same ChatGPT account on the mobile app². From there, the ChatGPT mobile app loads the live state of the connected environment. You can review outputs, approve commands, switch models mid-task, answer Codex's questions, change directions, and start new tasks — all without sitting at the keyboard where the agent is actually running. The phone receives live diffs, terminal output, screenshots, test results, and approval prompts as they happen².

A meaningful limit at launch: the mobile app only connects to the macOS version of the Codex desktop app. If your development machine runs Windows, you cannot drive Codex from your phone today even though Codex itself runs on Windows; OpenAI says Windows host support is coming. The Mac also has to stay awake, online, and signed in — if the host sleeps or the Codex app crashes, the mobile session ends².

Three companion updates landed in the same release:

HIPAA-compliant usage is still limited to eligible ChatGPT Enterprise workspaces running Codex in local environments².

The Growth Curve Behind the Release

Codex's user numbers explain why mobile, Chrome, and SSH all shipped in a single month. OpenAI says Codex now serves more than 4 million weekly users — a figure it crossed by April 21, 2026, after passing 3 million two weeks earlier and 2 million in March³. Inside ChatGPT Business and Enterprise specifically, OpenAI says weekly Codex usage grew about 6x between January and April 2026, and total Codex usage is up 10x since August 2025³.

At that adoption rate, the bottleneck is no longer model quality or context window — it is surface area. A developer who only meets Codex at a desktop is a part-time customer. Codex on your phone, in your browser, and over your SSH tunnels is a full-time one.

Codex in Chrome — A Week Earlier

The mobile release follows the Codex Chrome extension that OpenAI launched on May 7, 2026 and rolled out everywhere except the EU and UK on May 8⁴. The extension lets Codex use your signed-in browser to test web apps, pull context across tabs, drive DevTools, and operate against internal tools or sites like Gmail. It runs in the background in parallel across tabs and asks before touching any new website by default; an allowlist and blocklist live in the Computer Use settings⁵.

Chrome and mobile are complementary. Chrome gives Codex hands inside the browser; mobile gives the developer hands inside Codex from anywhere else.

How It Compares: Codex vs. Claude Code vs. Cursor

OpenAI is not first to mobile. Anthropic shipped Claude Code on iOS and the web as a research preview in October 2025 for Pro and Max subscribers, then released the official Claude Android app in March 2026, and rolled out "Remote Control" — a synchronization layer that lets users drive a local Claude Code CLI from a phone or tablet — on February 25, 2026, initially for Max only and now in research preview on both Pro ($20/month) and Max ($100 or $200/month) plans⁶. Cursor's path is different: cloud agents launched on February 24, 2026, with a third-party "Cursor AI Mobile: Remote IDE" app on the iOS App Store and a fully functional web dashboard at cursor.com/agents on Android⁷.

The free-tier point is the loudest difference. Anyone with a ChatGPT account — including Free — can drive Codex from a phone, provided their development machine is a Mac. Claude Code Remote Control starts at $20/month on Pro (research preview) and works regardless of host OS, but is paid-only. Cursor's mobile story still depends on a third-party iOS app or a web dashboard. None of this makes one tool better than another; it does shift who can adopt mobile coding agents at all, and on what hardware.

What Mobile Coding Agents Actually Change

The honest version is that mobile is not a place to write code. Tapping on a glass keyboard is still the wrong way to refactor a service. What mobile changes is the waiting — the part of agentic coding where the model is busy and the human is bored.

Three workflows benefit most:

The first is asynchronous review. You hand Codex a multi-file refactor on Friday afternoon, leave the office, and approve the diff from a phone before dinner. The phone shows the same terminal output and test results you would see at the desk².

The second is unblocking. Codex agents that run for hours occasionally need a small human decision — a credential, a yes/no on a destructive command, a choice between two paths. With mobile, the unblock happens in seconds rather than after a commute.

The third is monitoring long jobs. Codex SSH'd into a build server and a 90-minute integration test suite are no longer two reasons to stay at the keyboard.

What mobile does not change: the agent still runs where the code lives, model quality is still the bottleneck for hard tasks, and the access-control surface area just got bigger. Programmatic access tokens are restricted to Business and Enterprise plans for exactly this reason².

Bottom Line

May 14, 2026 was not a model day. No new GPT, no new benchmark scores, no new pricing per million tokens. What changed was the shape of where Codex runs — phone, browser, SSH tunnel — and who can use it. The 2M-to-4M ramp from March through April pushed OpenAI to make coding agents portable, and the company shipped the surfaces in roughly the order developers asked for: Chrome on May 7, mobile on May 14.

The competitive picture is now clear. Anthropic, OpenAI, and Cursor all believe mobile is part of the coding-agent stack. The disagreement is who pays — and on which host. OpenAI's bet is that putting Codex on every plan, including Free, will compound the same growth curve that added a million weekly users in roughly two weeks during April. If you want a related read, the Claude Opus 4.7 benchmarks post covers the model side of the same race, and the Claude Code Agent Mode deep dive walks through how Anthropic shaped its own coding agent.