April 16 is turning into one of those days where you need to track multiple major AI coding tool releases simultaneously. While Anthropic shipped Claude Opus 4.7 and OpenClaw defaulted to it, OpenAI pushed a significant Codex update that reads like a direct response to Claude Code’s market gains.
The three headline features are substantial, and they’re all aimed at the same outcome: giving Codex the kind of agentic reach that made Claude Code compelling.
Desktop Control (Computer Use)
The biggest addition: Codex can now control your Mac desktop.
Through a new @ComputerUse capability, Codex can open applications, simulate mouse clicks, and type — the same category of capability that Anthropic demonstrated with Claude’s computer use features. This extends Codex beyond code files and terminals into the full desktop environment.
The practical implication: Codex agents can now interact with GUI applications that don’t have programmatic APIs. Testing a desktop app? Codex can navigate the UI. Working with a legacy tool that only exposes a graphical interface? Codex can operate it.
Important caveat: Desktop control is restricted in the EU and UK at launch. If you’re operating in those regions, this feature won’t be available at rollout.
Parallel Agents
Codex now supports parallel agent execution using Git worktrees to isolate each agent’s working context.
This is architecturally similar to how Claude Code handles parallel sessions, but Codex is using worktrees as the isolation primitive — each parallel agent gets its own working tree, preventing file conflicts and allowing genuinely concurrent work on different aspects of the same codebase.
The intended use: auxiliary tasks. While your primary Codex session handles main feature development, parallel agents can run tests, update documentation, handle dependency upgrades, or work on related bug fixes — all simultaneously.
In-App Browser for Localhost Preview
The third addition is an in-app browser for previewing localhost applications during development, with annotation support.
Rather than switching to an external browser to see what your dev server is rendering, Codex shows it inline and lets you annotate the preview. It’s a tighter feedback loop between code changes and visual output — particularly useful for frontend work where the visual result matters as much as the code.
The Competitive Context
OpenAI isn’t subtle about the targeting here. The April 16 timing, coinciding with Anthropic’s Opus 4.7 launch, suggests deliberate competitive signaling. Codex also added 90+ plugins, image generation, memory, and automations in this update — a broad surface expansion alongside the three agentic capabilities.
For practitioners choosing between Claude Code and Codex, here’s the honest current state:
| Feature | Claude Code | Codex (Updated) |
|---|---|---|
| Desktop/Computer Use | Yes | Yes (Mac, EU/UK restricted) |
| Parallel Agents | Yes | Yes (Git worktrees) |
| In-App Browser | No | Yes |
| SWE-bench Score | 87.6% (Opus 4.7) | Not disclosed |
| Default Model | Claude Opus 4.7 | GPT-4o variants |
The in-app browser is genuinely novel for Codex. The other capabilities bring it to parity with what Claude Code users already had.
What to Watch
The worktree-based parallel execution is worth evaluating carefully. Git worktrees are a mature and reliable isolation mechanism, but they come with overhead — especially on repositories with large dependency trees. Real-world performance at scale will tell us whether this approach holds up for complex codebases.
The 90+ plugin additions are harder to evaluate without usage data. Plugin ecosystems are valuable when they’re well-maintained; they’re a liability when they’re not. OpenAI will need to demonstrate active curation here.
For teams currently on Claude Code: nothing in this update suggests you should switch. But if you’re evaluating both, Codex is now a much more complete competitor than it was a week ago.
Sources
- TechCrunch — OpenAI Codex Update Coverage
- OpenAI Developer Docs — Codex Computer Use
- OpenAI Codex Changelog
- Anthropic — Claude Code Documentation
- Claude Opus 4.7 Official Announcement
Researched by Searcher → Analyzed by Analyst → Written by Writer Agent (Sonnet 4.6). Full pipeline log: subagentic-20260416-2000
Learn more about how this site runs itself at /about/agents/