What happened

OpenAI shipped Codex 0.148.0 on August 18. Codex is OpenAI’s command-line coding agent: you run it in a terminal and it reads, writes, and executes code on your machine, similar to Claude Code.

Two changes stand out. First, you can now fork a session with codex exec fork, meaning Codex clones your current conversation and its context into a new, independent copy instead of continuing to edit the same one. You can also archive or restore past sessions from the TUI (the terminal interface Codex runs in) resume picker.

Second, Codex can now show “estimated thread credits or cost” directly in /status, in the terminal’s status line, and in the terminal window’s title bar. Credits are OpenAI’s unit for metering how much of a plan’s usage allowance a session has consumed, so this is a running estimate of what the current session is using or costing. The release notes say this only applies to “eligible workspaces,” and OpenAI hasn’t published what makes a workspace eligible or how many users that covers.

Why it matters

Forking solves a problem anyone using a coding agent for more than small edits has run into. Say Codex has spent twenty minutes building context on your codebase, and you want to try a riskier approach to the same task without losing that progress if it goes wrong. Before this release, your options were to keep editing the same session and hope you could undo a bad turn, or start a fresh session and rebuild that context from scratch. Forking gives you a third option: branch off, try the risky version, and the original session is still there if the branch doesn’t work out. It’s the same idea coding tools increasingly offer at the whole-app level, not just inside a single session, as with Base44’s recent Branches feature for testing app changes without touching the live version.

The cost visibility feature addresses a different, real pain point: coding agents can run for a while and rack up spend you don’t see until the bill arrives. Showing an estimate in the terminal title, somewhere you’re already looking, is a genuinely useful place to put it. The catch is that OpenAI gated it to “eligible workspaces” without saying which ones qualify. If you update Codex and don’t see a cost figure anywhere, that’s likely why, not a bug on your end.

Who should care

Anyone running longer Codex sessions where a wrong turn costs real rework time, and anyone who has been surprised by how much a session cost after the fact. If you’ve been running Codex without any sense of ongoing spend, this feature is worth checking for, even though it may not show up yet.

What builders should do next

To fork a session, run codex exec fork from your terminal while in an existing Codex session. That gives you a new session that starts from the same context but won’t affect the original if you keep working in it. Use it before trying an approach you’re not confident about, so backing out means switching sessions instead of undoing changes by hand.

To check whether you have cost visibility, look at your terminal’s title bar or window tab while running Codex, or type /status inside a session. If nothing shows up, your workspace likely isn’t in the group OpenAI enabled it for yet.

Update with npm install -g @openai/codex@latest if you installed through npm, or run codex update to use Codex’s built-in updater. Confirm you’re on the new version with codex --version.


End of article