Codex CLI 0.164.0 was published on 23 October 2026. This article covers that specific release, with an npm example pinned to its version. It is an application update; access to models and connected services still depends on your account and configuration.
1. What OpenAI released on 23 October
OpenAI published the Codex CLI 0.164.0 GitHub release at 04:48 UTC on 23 October 2026. The public npm registry records the matching @openai/codex package a few minutes later, at 04:52 UTC. Both records identify the same stable version; this article does not cover the separate alpha channel.
Codex CLI is OpenAI’s coding agent for the terminal. The release notes describe changes to the application: prompt reuse, interactive controls, enterprise connections, and reliability. They do not announce a new model or establish that generated code is more accurate.
For developers working on websites, custom software, or ERP modules, the useful question is which repeated task becomes easier to run and verify. Start with one small workflow before changing a whole team’s setup.
2. The changes that matter in daily work
On small screens, scroll the table sideways to see every column.
| Change | What it means for your workflow |
|---|---|
Reusable /presets | Ten numbered slots hold prompts or slash commands, with a keyboard menu for quick access. Useful for recurring repository reviews or handover summaries. |
| Inline model and permission controls | /model accepts model, reasoning, and speed arguments; /permissions accepts permission profiles. Choose values supported by your setup. |
| Images in the fullscreen transcript | Viewed and generated images can be previewed on terminals supporting Kitty or Sixel graphics. Availability depends on the terminal. |
| Enterprise MCP authentication | Supported connections can use eligible Codex account grants or dedicated sign-in, including OAuth clients that require secrets. Server and organization requirements still apply. |
| Credential masking | Masking is added for supported remote Bash commands and Zsh snapshot sessions, including configured credential providers. This does not mean every possible secret is detected. |
| Session and Windows connection fixes | Resume can find compressed histories and restore saved thread settings. Windows HTTPS and WebSocket proxy challenges can use the signed-in user’s Negotiate or NTLM credentials. |
OpenAI also reports fixes for failed conversation compaction, pending permission selections, and sandbox checks. These are release-note claims, rather than results from an independent benchmark. There is no measured productivity or accuracy improvement claimed here.
3. Install or update the npm version
Use this route if you already use npm for Codex, or have a supported development computer with a maintained Node.js release and npm installed. For a Homebrew or standalone installation, keep using that installation method; mixing methods can leave multiple executables on your PATH.
First, check Node.js and npm in your terminal. If either command is unavailable, finish setting up that toolchain before using the npm command below.
node --version
npm --versionThis installs the exact version discussed in the article. If you already run a newer version, keep it unless you have a specific reason to reproduce this older release. Do not downgrade a working installation just to follow the example.
npm install -g @openai/codex@0.164.0
codex --versionThe version check should report 0.164.0. If it reports another version, check which installation your terminal resolves and use the matching package manager. Do not fix a PATH problem by disabling the application’s permissions.
Open a terminal in a project you are authorized to work on, preserve your current work, and start the application:
codexComplete the sign-in flow offered by your installation. Installing the CLI does not include a subscription, API credit, or access to an enterprise MCP server. Review the active model and permissions before starting a task.
4. Try a reusable repository-review preset
A preset is useful when the instruction stays similar while the project changes. Start with an inspection task whose output you can compare against real files. Inside the interactive Codex session, open the preset controls; this is not a Bash or PowerShell command:
/presetsUse the preset editor to save a prompt such as the following in an empty numbered slot. Give it a recognizable label, such as “Repository overview.” Inspect an occupied slot before replacing it.
Inspect this repository and summarize its entry points, development commands, and tests. Support each finding with file paths. Separate confirmed facts from missing information. Do not install dependencies, change files, or call external services for this task.Recall the saved preset from the keyboard menu and check its response against the README, manifests, and named files. For an incomplete project, a useful answer identifies what is missing. A confident description of a nonexistent database or service is a failed check.
The prompt describes the task boundary; enforce the actual access boundary with permission settings. Presets save inputs. A reusable skill can contain longer instructions and supporting files, while an MCP server connects tools or data. Choose the mechanism that fits the work.
5. Check your setup after updating
- Model and permissions: open
/modeland/permissionsto inspect the choices available to your session. Review selected values before saving a command preset. - Saved sessions: reopen a non-sensitive practice session and confirm that its expected settings return. The release notes describe resume fixes; still check your own workflow.
- Connected tools: for MCP, complete the supported sign-in and test a read-only operation first. Authentication does not grant access to every business record or authorize every write.
- Images and proxies: check image preview only on a compatible terminal. If you use a Windows corporate proxy, confirm connectivity under your organization’s supported configuration.
Compare an existing task before and after the update using the same repository revision and expected output. Record the CLI version, selected model, and relevant settings. This helps distinguish an application change from a model, account, or project change.
6. Where this helps software and ERP teams
A web team could save a prompt for reviewing a small change before handover. An Odoo team could reuse an instruction to map a module’s manifest, models, views, and tests. An integration team could check a connected read operation before allowing actions that affect customers or stock.
These are suggested uses, not automatic capabilities delivered by the update. The application still needs the right project context, available tools, and clear acceptance criteria. Keep final decisions tied to working code, tests, and a human review appropriate to the change.
Try one preset with a clear output, confirm it works on a representative task, and document the version and settings your team adopts. That produces a repeatable workflow you can evaluate when the next release arrives.
7. Frequently asked questions
Is Codex 0.164.0 a new AI model?
No. It is a Codex CLI application release published on 23 October 2026. Model availability and behavior depend on the model selected and the access provided by your account.
Are Codex presets the same as skills or MCP?
No. Presets save prompts or slash commands for reuse. Skills can package instructions and supporting files. MCP connects an AI host to server capabilities such as tools and resources; it has separate setup and access requirements.
Does installing the update make Codex free to use?
Installing the CLI does not include a subscription or API credit. Sign-in, usage limits, billing, and enterprise connections depend on your account and the services you use.