Image: OpenAI
Why it mattersThe model picker lets you try a new model on the same task without losing the old chat. The side chat lets you run two jobs in one window.
A coding agent you use every day is held together by the small things. The model it picks by default. How you tell it which model to use for one chat. And whether a second question kicks the first one off its rails. OpenAI's Codex has a new release that touches each of those. Version 0.168.0 is out on the project's GitHub.
The release notes list seven new features and six bug fixes. Two of the features matter most for daily work. The rest are small changes a daily user will notice.
The model provider picker
OpenAI says you can now browse model providers under Sessions, then Model providers, and pick a default for new chats. Every new chat uses that default. An open chat keeps the provider you picked for it when you later change the default. A resumed chat uses its saved provider too.
That is the small change that takes the trial-and-error out of trying a new model on the same task. A chat you were running on one model stays on it. A new chat starts on the one you switched to. Nothing else in your work has to change.
The workspace side chat
OpenAI calls the second feature a workspace. You turn it on by setting features.tui_workspace. A side chat opens next to the main one. You drag a divider to resize the panes. You press leader z to zoom the focused pane. The side chat runs its own session. The main chat does not pause while you use the side one.
That is the part that opens a different shape of work for a Codex user. A long main job stays open. A short side job runs next to it. You switch between them without closing either. The project's own changelog links four pull requests that built this feature.
The smaller features
The notes list more that is worth knowing about. Codex can attach audio recordings when you talk to a model that can hear. It skips a recording you have marked as rejected on later turns. It uses the clip length to keep more short recordings in context. The release notes link two pull requests for that work.
A new image-generation flag lets Codex make animated GIFs. You set a frame count and a timing. On the macOS terminals Ghostty and Kitty, Codex can now draw image previews in tmux. You need the allow-passthrough all flag for that. Terminal notifications can use OSC 777. Approval prompts show longer command previews. They also show the difference between terminal input and a network call.
What a Codex user can act on
If you use Codex every day, the two features at the top of this item are the ones to try. The model picker is set under Sessions, then Model providers. The workspace is behind the features.tui_workspace flag. The audio attachment is on by default for any model that supports it.
The release was published to the openai/codex repository, and the notes list 171 pull requests in all. Nothing here breaks a workflow. The point is a more controllable daily tool.
Source
- Codex 0.168.0 release notes, OpenAI
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
