Codex CLI: what’s new in 0.164.0 and our verdict
OpenAI's open-source coding agent that runs in your terminal.
At a glance
- Evidence
- Tested (smoke test)
- Released
- Oct 23, 2026
- Last verified
- Oct 25, 2026
- Pricing
- Not verified
- Best for
- AI coding agents, vibe coding
Evidence and ranking
Evidence behind this verdict
- Sources read2 cited sources, latest fetched Oct 25, 2026. (done)
- Facts re-checkedLast verified Oct 25, 2026. (done)
- Run by usSmoke test on Oct 25, 2026 (scope below). (done)
Partly worked Test log result: Partly worked. Read the log
Where we rank it (editorial)
- AI coding agents#2 of 20 tools
- vibe coding#2 of 11 tools
Editorial picks, not benchmark scores. How we rank.
Pricing: We have not verified pricing for this tool; see the official site.
Our verdict
A quality-of-life release: worth updating if you re-type the same prompts or switch models a lot, not a reason to switch tools.
Good for
- People who already use Codex and repeat the same setup prompts (presets save the typing)
- Teams that connect Codex to enterprise MCP servers and want fewer sign-in hurdles
- Anyone who wants an open-source (Apache-2.0) agent they can run from a terminal
Skip if
- You are happy with another coding agent; nothing here changes the core capability
- You need the image preview feature but your terminal does not support Kitty or Sixel graphics
Tested. Tested 2026-10-25 (America/Chicago). Installed @openai/codex from npm: codex-cli 0.164.0. Ran "codex exec --oss --local-provider ollama -s workspace-write --json" against a local Ollama model (no ChatGPT sign-in; the model name needs the :latest tag). Told to run a sed command, Codex executed it in its sandbox (exit code 0) and calc.py changed a-b to a+b. A plain-language edit request ran no edit (the small model tried a sub-agent call that failed to parse); we kept two runs, so this is not a pass rate. Not tested: sign-in, presets, /model and /permissions arguments, MCP sign-in or image previews. Test environment: Linux x86_64 box with 8 CPUs and no GPU, no vendor account or API key. Agent tools were pointed at a local open-weight model (Qwen2.5-3B-Instruct, 4-bit, served by Ollama 0.40.11), so we tested the tool, not any vendor's hosted models. Full log: /tests/codex-cli.
Full test log with exact commands, versions and raw output: Codex CLI 0.164.0: our test log.
- Evidence
- Tested (smoke test, 2026-10-25, local Qwen2.5-3B-Instruct, not vendor models). A smoke test is not a benchmark.
- What we ran
- Installed @openai/codex from npm (codex-cli 0.164.0) and ran 'codex exec --oss' against a local model; Codex executed a sed command in its sandbox and changed calc.py. A plain-language edit request ran no edit; we kept two runs.
- Model used
- Qwen2.5-3B-Instruct (4-bit), a small local open-weight model served by Ollama 0.40.11; no vendor-hosted model was used
- Environment
- Linux x86_64 box, 8 CPUs, no GPU; no vendor account, login or API key
- Not tested
- ChatGPT sign-in, presets, /model and /permissions arguments, MCP sign-in, image previews, OpenAI-hosted models.
- Date
- (America/Chicago)
Get the next tool verdict in your inbox. Free, about six minutes a week.
Free. Unsubscribe anytime. No spam. We email a confirmation link first. Privacy
Best for
- Best AI tool for AI coding agents: A mature open-source terminal agent that ships frequent updates; this one is about workflow polish.
- Best AI tool for vibe coding: Workable for vibe coding if you are comfortable in a terminal; the presets slots help repeat your favorite starter prompts.
What’s new
Codex CLI 0.164.0 adds numbered /presets slots for reusable prompts, inline arguments for /model and /permissions, image previews in supporting terminals, enterprise MCP sign-in and credential masking.
- Save reusable prompts and slash commands in ten numbered /presets slots, with a keyboard menu for quick access.
- Set model, reasoning effort, speed and permission profiles directly with inline /model and /permissions arguments.
- Preview viewed and generated images in the fullscreen transcript on terminals that support Kitty or Sixel graphics.
- Authenticate enterprise MCP connections using eligible Codex account grants or dedicated sign-in, including OAuth clients that require secrets.
- Mask credentials during supported remote Bash commands and Zsh snapshot sessions, including credentials from custom providers.
- The Amazon Bedrock setup wizard is now enabled by default.
- Bug fixes: failed compaction no longer leaves partial summaries in history, and resumed sessions restore saved thread settings instead of global defaults.
How to set it up: Install Codex CLI and try the new presets
Install steps come from the project README. We ran the install and a short sandboxed task on 2026-10-25. The presets steps are based on the release notes only and were not tested; the notes do not document the exact save flow, so we have not written steps for it.
Install Codex CLI
On Mac or Linux, run the official installer. On Windows use the PowerShell installer in the README. npm and Homebrew are also supported.
curl -fsSL https://chatgpt.com/codex/install.sh | sh # or: npm install -g @openai/codex # or: brew install --cask codexStart it and sign in
Run codex and choose Sign in with ChatGPT to use it with your ChatGPT plan (the README lists Plus, Pro, Business, Edu and Enterprise).
codexCheck you are on 0.164.0 or newer
The features below shipped in 0.164.0. If you installed from a package manager, update it first.
Open the presets menu
The release notes say reusable prompts and slash commands can be saved in ten numbered /presets slots with a keyboard menu. Type /presets inside Codex and follow the on-screen prompts. (Not tested by us; the exact save flow is not in the release notes.)
/presetsUse inline arguments on /model and /permissions
Per the release notes, /model and /permissions now accept arguments inline so you can set model, reasoning effort, speed and permission profile in one command. The argument syntax is not in the notes; run /model and /permissions to see the options. (Not tested by us.)
Pros and cons
Pros
- Open source under Apache-2.0 (per the GitHub repository metadata)
- Installs with a one-line script, npm or Homebrew
- Frequent releases with detailed changelogs
- Credential masking and sandbox hardening continue to improve
Cons
- Release notes do not explain how to use several new features; expect to read the in-app help
- Image previews need a Kitty- or Sixel-capable terminal
- Alpha builds ship almost daily, so make sure you install a stable tag
- Our test confirmed the install and sandboxed command execution only; with a small local model, plain-language edits were unreliable, and the new presets were not tested
Alternatives and comparisons
- Claude Code: Anthropic's agentic coding tool for your terminal, IDE and GitHub.
- Gemini CLI: Google's open-source terminal agent for Gemini, now largely replaced by Antigravity CLI for consumers.
- Kiro: AWS's agentic coding tool that turns prompts into requirements, designs and tasks (spec mode), with an IDE and a CLI.
- Cursor: AI code editor and agent platform with cloud agents, a CLI and its own Git hosting (Origin).
OpenAI models we track
- GPT-6.1 Terra (released Oct 26, 2026)
- GPT-6.1 Sol (released Sep 29, 2026)
- GPT-6 Luna (released Sep 22, 2026)
- GPT-6 Sol (released Sep 22, 2026)
- GPT-Live 1 (released Sep 10, 2026)
- GPT-Image-2.5 Flare (released Sep 8, 2026)
FAQ
What is new in Codex CLI 0.164.0?
Numbered /presets slots for reusable prompts, inline arguments for /model and /permissions, image previews on Kitty or Sixel terminals, enterprise MCP sign-in, credential masking in remote Bash sessions and the Amazon Bedrock setup wizard enabled by default.
How do I install Codex CLI?
On Mac or Linux run curl -fsSL https://chatgpt.com/codex/install.sh | sh. You can also use npm install -g @openai/codex or brew install --cask codex.
Is Codex CLI free and open source?
The code is open source under Apache-2.0. Using it with your ChatGPT plan depends on that plan; check OpenAI's pricing for current limits.
What we have not verified
- How to save a preset and the exact /model and /permissions argument syntax (not documented in the release notes)
- Pricing and plan limits for using Codex (not fetched)
- Presets, /model and /permissions arguments, MCP sign-in and image previews (not tested; they need an interactive, signed-in session)
Sources
- GitHub release rust-v0.164.0 (fetched 2026-10-25)
- Codex README (install instructions) (fetched 2026-10-25)
Last verified · Agentic Weekly
Part of the week of Oct 19, 2026 archive.
Guides for Codex CLI
- ChatGPT Work vs ChatGPT Chat vs Codex: what each is for (Reported (not run by us): from OpenAI's documentation; we have not used ChatGPT Work; last verified Oct 26, 2026)
- Which coding agents can use GPT-6.1 Terra? (Reported (not run by us): from vendor docs and changelogs; last verified Oct 26, 2026)
Latest news about Codex CLI
- OpenAI GPT-6.1 Terra: a $1/$5 computer-use model, and what the launch post leaves out (Reported; first seen 2026-10-26 19:05 UTC)
- Superset is now Paradev: parallel coding-agent app ships desktop v2.0.0 and an $11.5M seed (Reported; first seen 2026-10-26 17:05 UTC)
- Octopus SI OS: an open-source, self-evolving agentic OS with a Rust kernel, released as a research preview (Reported; first seen 2026-10-26 12:27 UTC)
Related tools
- Kiro: AWS's agentic coding tool that turns prompts into requirements, designs and tasks (spec mode), with an IDE and a CLI.
- Claude Code: Update now if you use the sandbox or run Claude Code on a cloud provider; skip if you wanted flashy new features, this is a hardening release.
- Cursor: Worth a look if you run agents on several repos or want one place for editor, cloud agents and Git hosting; the October changes are mostly for teams.
How to cite this page
Agentic Weekly, "Codex CLI 0.164.0: what's new, verdict & setup", last updated 2026-10-25, https://theagenticweekly.com/tools/codex-cli
Keep the Reported or Tested label when you quote a verdict. Plain-text version: https://theagenticweekly.com/tools/codex-cli.md. More for AI assistants: /ai, /tools.json, /llms.txt.
Free · weekly · about six minutes
Want the next verdict in your inbox?
What shipped this week, what is worth switching to, and what to ignore. We run what we can and label what we only read.
- One email a week. Each release gets one call: update now, worth a look, or skip.
- Every claim shows its evidence. Reported means we read the sources; Tested means we ran it, with the exact scope and the failures left in.
- A free AI Tools Cheat Sheet (web page and printable PDF) lands in your inbox when you confirm.
Free. Unsubscribe anytime. No spam. We email a confirmation link first and add you only after you click it. We use your email only to send this newsletter and never sell it. Privacy