Key Takeaways

• OpenAI and Anthropic shipped competing upgrades within two days of each other in April 2026 — Claude Code’s desktop redesign on April 14, Codex’s “Codex for (almost) everything” on April 16.
• Codex moved beyond coding into a full desktop agent: it can control macOS apps, browse the web, generate images, run scheduled tasks, and connect to 90+ plugins. At launch on April 16, 2026, computer use was macOS-only.
• Claude Code’s redesign added multi-session windows, a drag-and-drop workspace, an integrated terminal and file editor, plus Routines — cloud automations triggered by schedule, API call, or GitHub events.
• Codex now serves 3 million weekly developer users; within ChatGPT Business and Enterprise, Codex users grew 6x between January and April 2026.
• GPT-5.3-Codex launched February 5, 2026, roughly 25% faster than its predecessor, and lets users steer the model in real time while it works.
• Neither tool is a clean winner: Codex bets on breadth (one app for everything), Claude Code bets on depth (parallel coding sessions and cloud automation for engineers).

Codex vs Claude Code: Which Is Ahead Right Now?

As of late May 2026, neither tool holds a decisive lead — they are pulling in different directions. Codex has stopped being a coding assistant and turned into a general desktop agent that happens to write code. Claude Code has doubled down on serious engineering work, adding parallel sessions and server-side automation aimed squarely at professional developers. If you want one app that controls your Mac, fills your browser, and runs background errands, Codex is in front. If you want to run several coding tasks at once and wire automated jobs into your repositories, Claude Code leads.

The reason the question is hard to answer is timing. Both companies upgraded within 48 hours of each other in mid-April 2026, and each move read partly as a reply to the other. OpenAI’s update landed the same week Anthropic redesigned Claude Code’s desktop app with multi-sessions and Routines. The two products now overlap less than they used to, which makes the race less about a single benchmark and more about which philosophy fits your work.

What Changed in Codex: From Code Editor to Desktop Agent

OpenAI released its “Codex for (almost) everything” update on April 16, 2026. It is not a new product — it is a repositioning of an existing one, turning Codex from a coding agent into a full developer workstation that happens to run AI. The headline addition is computer use. Codex can now operate any macOS app — Figma, Xcode, Slack, the browser — by seeing the screen, moving a cursor, clicking, and typing, while you keep working in other apps.

The update also brought a native browser, image generation, a memory preview, more than 90 plugins, and extended automations. OpenAI framed the package as Codex 2.0 rather than a fresh app. The scale behind the move is real: three million developers used Codex in the week of the launch, and inside ChatGPT Business and Enterprise the user count grew sixfold between January and April 2026, with Notion, Ramp, and Braintrust among named customers.

The Native Browser and Its Limits

The in-app browser is the most visible new feature, but it ships with deliberate restrictions. The native Codex browser does not support cookies, user profiles, extensions, or pages requiring login. The reasoning is security: if a page displays a value the user controls, Codex could receive unexpected instructions, so OpenAI advises against mixing credentials with browser-assisted workflows. To test authenticated flows, the recommended path is a local mock or an unprotected staging environment, never production credentials. It is a trade-off that raises safety at the cost of some of the firepower shown in demos.

GPT-5.3-Codex and Real-Time Steering

The model under the hood also advanced. OpenAI launched GPT-5.3-Codex on February 5, 2026, calling it its most capable agentic coding model — the first to combine the Codex and GPT-5 training stacks, about 25% faster than its predecessor. The standout behavior is interaction. With the new model, Codex sends frequent updates so you stay aware of key decisions; instead of waiting for a final output, you can ask questions, discuss approaches, and steer toward the solution in real time. On benchmarks, OpenAI reported 77.3% on Terminal-Bench 2.0 and 64.7% on OSWorld-Verified, with the SWE-Bench Pro gain more incremental at 56.8% versus 56.4% for GPT-5.2-Codex.

What Changed in Claude Code: Parallel Work and Cloud Routines

Anthropic shipped its redesign two days earlier, on April 14, 2026. It was the most significant interface update Claude Code had received since its original CLI launch, adding a multi-session sidebar, a drag-and-drop workspace layout, an integrated terminal and file editor, Git worktree isolation, three view modes, side chat, and cloud-based Routines. The goal was practical: run more Claude Code tasks at once.

Before the change, juggling multiple sessions meant separate terminal windows with no unified view. One developer described running four Claude Code sessions across four terminal windows for most of 2026 — backend work on a client SaaS, a Next.js frontend rebuild, a content-research scraper, and a rotating experiment. The new sidebar collapses that into one window with status filters and logical grouping.

Routines: Automation That Runs Without You

The most consequential addition is Routines. These are saved, autonomous Claude Code tasks that run on Anthropic’s cloud infrastructure rather than your local machine — configure once, then trigger by schedule, API call, or GitHub events. You give Claude a prompt and a cadence, and it runs on that schedule: for example, every night at 2am, pull the top bug from Linear, attempt a fix, and open a draft PR.

The API trigger is where Routines stretch beyond coding. Each routine gets a unique HTTP endpoint and a bearer token; you POST to the endpoint with an optional text payload and the routine fires. You can wire Claude Code into alerting, deploy hooks, or internal tools — read an alert payload, find the owning service, and post a triage summary with a proposed first step. The GitHub trigger goes further still: subscribe a routine to repository events, and Claude creates a new session for every PR matching your filters and runs your routine.

Interface Controls for Power Users

The redesign added depth controls too. Three view modes — Verbose, Normal, and Summary — let you dial the interface from full transparency into Claude’s tool calls down to just the results. New keyboard shortcuts cover session switching, spawning, and navigation, a usage button shows context window and session usage at a glance, and the app now streams responses as Claude generates them. The redesigned desktop app is available to all Claude Code users on Pro, Max, Team, and Enterprise plans, and through the Claude API. At the Code w/ Claude event in May, Anthropic also announced rate-limit increases, doubling the Claude Code five-hour limit for Pro, Max, and Enterprise customers.

Codex vs Claude Code: Feature Comparison

Model and Benchmark Snapshot

A note on the speed figure: independent reporting cites GPT-5.3-Codex as roughly 25% faster than Claude Opus 4.6 on technical benchmarks, though benchmark comparisons across vendors should be read with care, since each company tests on the suites that favor its design.

Which One Should You Choose?

The honest answer is that the two tools have stopped competing for the same user. Codex suits people who want a single application to operate their machine, browse, generate visuals, and automate errands — a broad assistant that codes among other things. Claude Code suits engineers who want to run several focused coding tasks in parallel and push automated jobs into their repositories and alerting systems. Together, the redesign and Routines shift Claude Code from “AI pair programmer” toward something closer to an AI operations platform.

Both companies are iterating fast, and the gap will likely narrow again. OpenAI watched Claude Code take market share for six months and responded with a product that goes far beyond code editing. Anthropic answered with parallelism and cloud automation. For now, the lead belongs to whichever fits your daily work — not to a scoreboard.

If you are interested in this topic, we suggest you check our articles:

• What Tasks Are Best for Claude Code or OpenAI Codex?
• Which AI LLM is Best for Vibe Coding?

Sources: OpenAI – GPT-5.3-Codex · OpenAI Model Release Notes · BuildFastWithAI – Codex for (almost) everything · BuildFastWithAI – Claude Code Desktop Redesign · Releasebot – Claude Updates · DigitalApplied – GPT-5.3-Codex Benchmarks · Wikipedia – GPT-5.3-Codex

Written by Alius Noreika