OpenAI just dropped a dedicated macOS app for Codex, and it’s basically a command center for anyone who wants to stop writing code and start supervising it.
Launched Monday, Feb 2, the app is a direct shot at Anthropic’s Claude Code. The big flex? It lets you run multiple AI agents in parallel, each grinding away on different tasks while you (theoretically) do something more important.
Now this isn’t just another chatbot window. OpenAI is pushing "agentic coding," where the AI doesn't just suggest a line of code; it builds the whole feature. We’re talking:
Multi-Agent Parallelism: You can fire up separate threads for different tasks (one for bug hunting, one for refactoring, one for that feature you’ve been procrastinating on).
Isolated Worktrees: Every agent gets its own "sandbox" copy of your code. So no stepping on toes, or messy merge conflicts.
Agent Personality Selection: You can actually choose your agent’s "vibe." If you want a no-nonsense, pragmatic partner, or prefer something empathetic and collaborative, you got it.
Skills System: Codex isn’t stuck just writing code anymore. You can bundle instructions, tools, and scripts into skills, then choose them manually or let Codex auto-select what works best.
Background Automations: You can literally schedule Codex to "Review your codebase for security flaws at 2 AM every night." and wake up to a list of fixes. Sam Altman called it the "most loved internal product" at OpenAI.
Oh and get this: Each agent can work for up to 30 minutes independently!
The Benchmarks?
OpenAI is powering this with GPT-5.2-Codex, and the stats are... well, they're tight.
On TerminalBench: Codex-5.2 holds the top position. But in reality: It’s statistically tied with Claude Opus 4.5 and Gemini 3. So yeah, we've officially reached "margin of error" territory where the better UI wins, not just the smarter brain.
On SWE-bench Verified: The outcome is pretty much a tie also, at the end of the day, the performance gap between the top three models is so tight that user experience matters more than raw scores.
How To Get It?
It’s rolling out now for ChatGPT Plus, Pro, Enterprise, Edu and Business users.
Launch Perk: Paid users get 2x rate limits right now to celebrate the launch.
Freebie: Even Free and Go tier users can try it for a limited time.
The Bottom Line: If you’re a Mac-using developer, this is a massive upgrade over "copy-pasting into a browser." It’s OpenAI’s play to keep you from jumping ship to Anthropic—and with background automations and multi-agent threads, it might just work.
