Every hour your engineering team spends context-switching between a chat interface and a terminal is an hour they are not shipping.

The copy-paste loop between AI chat and your codebase just got replaced

Developers have been running AI suggestions in one window and pasting them into their editor in another, manually verifying each step. That two-window tax adds up fast on any team doing serious output.

One CLI command now puts a reasoning model directly inside your terminal

[AINews] OpenAI o3, o4 is an open-source command-line tool that takes natural language instructions as input and executes code changes, file edits, and shell commands directly in your working environment. You install it, point it at your project, and issue instructions in plain English. The output is actual code changes, not suggestions you have to implement yourself.

Engineering leads are the first to feel the gap closing

• Senior engineers who review AI-generated PRs and catch context errors: they get a tool that reads the full repo before generating, reducing garbage suggestions.
• DevOps engineers who script repetitive infrastructure changes: they get natural language automation that writes and runs the shell commands on the spot.
• Technical leads evaluating AI coding tools for their teams: they get an open-source option they can audit, fork, and deploy without a vendor contract.

The o3 and o4-mini model upgrades powering this release are built on 10x the compute applied to reinforcement learning, and OpenAI is explicit that vision and tool use improved most. That matters because Codex CLI is not just text generation, it is a tool-use chain running inside your actual file system.

Claude Code had a head start, and OpenAI just went open-source to counter it

• Run natural language refactors across an entire codebase in one command.
• Generate and execute shell scripts without leaving the terminal.
• Feed the model a bug report and get a tested patch written to disk.
• Fork and self-host the CLI since the full source is publicly available.

Pricing for o3 and o4-mini API access follows OpenAI’s standard tier structure; Codex CLI itself is free and open-source. Check our directory for current API rate details.

Tool-use capabilities are not yet available through the API, which means teams building on top of these models programmatically will wait before they see the full benefit.

Claude Code is the direct alternative if you want a managed experience with Anthropic’s models. For teams that want full control and open-source auditability, nothing else in this category currently matches Codex CLI on those terms.

The open-source move signals where the AI coding race is heading next

Proprietary coding assistants are losing ground to tools teams can inspect and own outright. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.