OpenAI · 2026-04-11 · notable
OpenAI Codex CLI v0.120 — Realtime V2 background streaming
OpenAI's terminal coding agent gains Realtime V2 streaming that surfaces background agent progress while work is still running, plus MCP structured tool results and sharper TUI hook displays in a 30-PR release.
OpenAI's terminal coding agent can now stream a background agent's progress live while it is still working.
Key specs
| Version | 0.120.0 |
|---|---|
| GitHub stars | 75.1k |
| Prs merged | 30 |
What is it?
OpenAI Codex CLI is OpenAI's official terminal-native coding agent — a Rust binary with 75.1k stars on GitHub that lives in your shell, reads and edits files, runs commands, and drives longer tasks through the Codex model family. v0.120.0 is a 30-PR point release layered on top of v0.119, which last week switched realtime voice sessions to WebRTC by default.
How does it work?
The headline change is Realtime V2 streaming: the background-agent pipeline now surfaces incremental progress while work is still in flight, and queues any follow-up user turns until the active response completes — no more 'running…' dead-time on a long task. Code-mode tools now consume MCP output schema details directly, so typed tool results flow end-to-end without shape guessing. The TUI splits hook activity into 'running' vs. 'finished output' rows, thread titles become addressable inside custom status-line templates, and SessionStart hooks can distinguish a /clear from a fresh start or a resumed session. The release also fixes Windows sandbox permission handling, symlink resolution, and remote websocket tool ordering.
Why does it matter?
Background streaming is what makes long-running agent work tolerable in a terminal: you can see progress instead of staring at a spinner, and you can already type the next instruction while the model is finishing. Combined with the v0.119 WebRTC voice stack, Codex CLI is catching up structurally with managed-agent harnesses from other labs for developers who want that experience in their own shell — without giving up local control.
Who is it for?
Devs who run coding agents from the terminal and want live background progress plus queued follow-ups.
Try it
npm install -g @openai/codex@0.120.0