OpenAI Codex CLI: Lightweight coding agent that runs in your terminal

I tried one task head-to-head with Codex o4-mini vs Claude Code: writing documentation for a tricky area of a medium-sized codebase.<p>Claude Code did great and wrote pretty decent docs.<p>Codex didn&#x27;t do well. It hallucinated a bunch of stuff that wasn&#x27;t in the code, and completely misrepresented the architecture - it started talking about server backends and REST APIs in an app that doesn&#x27;t have any of that.<p>I&#x27;m curious what went so wrong - feels like possibly an issue with loading in the right context and attending to it correctly? That seems like an area that Claude Code has really optimized for.<p>I have high hopes for o3 and o4-mini as models so I hope that other tests show better results! Also curious to see how Cursor etc. incorporate o3.

related demo&#x2F;intro video: <a href="https:&#x2F;&#x2F;x.com&#x2F;OpenAIDevs&#x2F;status&#x2F;1912556874211422572" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;OpenAIDevs&#x2F;status&#x2F;1912556874211422572</a><p>this is a direct answer to claude code which has been shipping furiously: <a href="https:&#x2F;&#x2F;x.com&#x2F;_catwu&#x2F;status&#x2F;1903130881205977320" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;_catwu&#x2F;status&#x2F;1903130881205977320</a><p>and is not open source; there are unverified comments that they have DMCA&#x27;ed decompilations <a href="https:&#x2F;&#x2F;x.com&#x2F;vikhyatk&#x2F;status&#x2F;1899997417736724858?s=46" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;vikhyatk&#x2F;status&#x2F;1899997417736724858?s=46</a><p>by total coincidence we&#x27;re releasing our claude code interview later this week that touches on a lot of these points + why code agent CLIs are an actually underrated point in the SWE design space<p>(TLDR you can use it like a linux utility - similar to @simonw&#x27;s `llm` - to sprinkle intelligence in all sorts of things like CI&#x2F;PR review without the overhead of buying a Devin or a Copilot SaaS)<p>if you are a Claude Code (and now OAI Codex) power user we want to hear use cases - CFP closing soon, apply here <a href="https:&#x2F;&#x2F;sessionize.com&#x2F;ai-engineer-worlds-fair-2025" rel="nofollow">https:&#x2F;&#x2F;sessionize.com&#x2F;ai-engineer-worlds-fair-2025</a>

These days, I usually paste my entire (or some) repo into gemini and then APPLY changes back into my code using this handy script i wrote: <a href="https:&#x2F;&#x2F;github.com&#x2F;asadm&#x2F;vibemode">https:&#x2F;&#x2F;github.com&#x2F;asadm&#x2F;vibemode</a><p>I have tried aider&#x2F;copilot&#x2F;continue&#x2F;etc. But they lack in one way or the other.

Fingers crossed for this to work well! Claude Code is pretty excellent.<p>I’m actually legitimately surprised how good it is, since other coding agents I’ve used before have mostly been a letdown, which made me only use Claude in direct change prompting with Zed (“implement xyz here”, “rewrite this function with abc”, etc), so very hands-on.<p>So I’ve went into trying out Claude Code rather pessimistically, and now I’m using it all the time! Sure, it ends up costing a bunch, but it’s easy to justify $15 for a prompting session if the end result is a mostly complete PR, done much faster.<p>All that is to say - competition is good, fingers crossed for codex!

This is pretty neat! I was able to use it for few use cases where it got it right the first time. The ability to use a screenshot to create an application is nice for rapid prototyping. And good to see them open sourcing it unlike claude.

First experience is not great. Here are the issues to start using codex:<p>1. Default model used doesn&#x27;t work and you get error: system OpenAI rejected the request (request ID: req_06727eaf1c5d1e3f900760d10ca565a7). Please verify your settings and try again.<p>2. You have to switch to model o4-mini-2025-04-16 or some other model using &#x2F;model. Now if you exit codex, you are back to default model and again have to switch everytime.<p>3. Crashed the first time with NodeJS error.<p>But after initial hickups seems to work and still checking how good&#x2F;bad it is compared to claude code (which I love except for context size limits)

Not sure why they used React for a CLI. The code in the repo feels like it was written by an LLM—too many inline comments. Interestingly, their agent&#x27;s system prompt mentions removing inline comments <a href="https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;utils&#x2F;agent&#x2F;agent-loop.ts#L1057">https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;util...</a>.<p>&gt; - Remove all inline comments you added as much as possible, even if they look normal. Check using \`git diff\`. Inline comments must be generally avoided, unless active maintainers of the repo, after long careful study of the code and the issue, will still misinterpret the code without the comments.

Strictly worse than Claude Code presently, but I hope since it&#x27;s open source that changes quickly.

<i>Next, set your OpenAI API key as an environment variable:</i><p><i>export OPENAI_API_KEY=&quot;your-api-key-here&quot;</i><p><i>Note: This command sets the key only for your current terminal session. To make it permanent, add the export line to your shell&#x27;s configuration file (e.g., ~&#x2F;.zshrc).</i><p>Can&#x27;t any 3rd party utility running in the same shell session phone home with the API key? I&#x27;d ideally want only codex to be able to access this var

Claude Code represents something far more than a coding capability to me. It can do anything a human can do within a terminal.<p>It’s exceptionally good at coding. Amazing software, really, I’m sure the cost hurdles will be resolved. Yet still often worth the spend

From my experience with playing with Claude Code vs Cline( which is open source and the tool to beat imo). I don&#x27;t want anything that doesn&#x27;t let me set my own models.<p>Deepseek is about 1&#x2F;20th of the price and only slightly behind Claude.<p>Both have a tendency to over engineer. It&#x27;s like a junior engineer who treats LOC as a KPI.

I&#x27;ve had great results with the Amazon Q developer cli, ever since it became agentic. I believe it&#x27;s using claude-3.7-sonnet under the hood.

Here is the prompt template, in case you&#x27;re interested:<p><pre><code> const prefix = `You are operating as and within the Codex CLI, a terminal-based agentic coding assistant built by OpenAI. It wraps OpenAI models to enable natural language interaction with a local codebase. You are expected to be precise, safe, and helpful. You can: - Receive user prompts, project context, and files. - Stream responses and emit function calls (e.g., shell commands, code edits). - Apply patches, run commands, and manage user approvals based on policy. - Work inside a sandboxed, git-backed workspace with rollback support. - Log telemetry so sessions can be replayed or inspected later. - More details on your functionality are available at \`codex --help\` The Codex CLI is open-sourced. Don&#x27;t confuse yourself with the old Codex language model built by OpenAI many moons ago (this is understandably top of mind for you!). Within this context, Codex refers to the open-source agentic coding interface. You are an agent - please keep going until the user&#x27;s query is completely resolved, before ending your turn and yielding back to the user. Only terminate your turn when you are sure that the problem is solved. If you are not sure about file content or codebase structure pertaining to the user&#x27;s request, use your tools to read files and gather the relevant information: do NOT guess or make up an answer. Please resolve the user&#x27;s task by editing and testing the code files in your current code execution session. You are a deployed coding agent. Your session allows for you to modify and run code. The repo(s) are already cloned in your working directory, and you must fully solve the problem for your answer to be considered correct. You MUST adhere to the following criteria when executing the task: - Working on the repo(s) in the current environment is allowed, even if they are proprietary. - Analyzing code for vulnerabilities is allowed. - Showing user code and tool call details is allowed. - User instructions may overwrite the *CODING GUIDELINES* section in this developer message. - Use \`apply_patch\` to edit files: {&quot;cmd&quot;:[&quot;apply_patch&quot;,&quot;*** Begin Patch\\n*** Update File: path&#x2F;to&#x2F;file.py\\n@@ def example():\\n- pass\\n+ return 123\\n*** End Patch&quot;]} - If completing the user&#x27;s task requires writing or modifying files: - Your code and final answer should follow these *CODING GUIDELINES*: - Fix the problem at the root cause rather than applying surface-level patches, when possible. - Avoid unneeded complexity in your solution. - Ignore unrelated bugs or broken tests; it is not your responsibility to fix them. - Update documentation as necessary. - Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task. - Use \`git log\` and \`git blame\` to search the history of the codebase if additional context is required; internet access is disabled. - NEVER add copyright or license headers unless specifically requested. - You do not need to \`git commit\` your changes; this will be done automatically for you. - If there is a .pre-commit-config.yaml, use \`pre-commit run --files ...\` to check that your changes pass the pre-commit checks. However, do not fix pre-existing errors on lines you didn&#x27;t touch. - If pre-commit doesn&#x27;t work after a few retries, politely inform the user that the pre-commit setup is broken. - Once you finish coding, you must - Check \`git status\` to sanity check your changes; revert any scratch files or changes. - Remove all inline comments you added much as possible, even if they look normal. Check using \`git diff\`. Inline comments must be generally avoided, unless active maintainers of the repo, after long careful study of the code and the issue, will still misinterpret the code without the comments. - Check if you accidentally add copyright or license headers. If so, remove them. - Try to run pre-commit if it is available. - For smaller tasks, describe in brief bullet points - For more complex tasks, include brief high-level description, use bullet points, and include details that would be relevant to a code reviewer. - If completing the user&#x27;s task DOES NOT require writing or modifying files (e.g., the user asks a question about the code base): - Respond in a friendly tune as a remote teammate, who is knowledgeable, capable and eager to help with coding. - When your task involves writing or modifying files: - Do NOT tell the user to &quot;save the file&quot; or &quot;copy the code into a file&quot; if you already created or modified the file using \`apply_patch\`. Instead, reference the file as already saved. - Do NOT show the full contents of large files you have already written, unless the user explicitly asks for them.`; </code></pre> <a href="https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;utils&#x2F;agent&#x2F;agent-loop.ts">https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;util...</a>

<a href="https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;components&#x2F;chat&#x2F;terminal-chat-input-thinking.tsx">https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;codex&#x2F;blob&#x2F;main&#x2F;codex-cli&#x2F;src&#x2F;comp...</a><p>Hey comment this thing in!<p><pre><code> const thinkingTexts = [&quot;Thinking&quot;]; &#x2F;* [ &quot;Consulting the rubber duck&quot;, &quot;Maximizing paperclips&quot;, &quot;Reticulating splines&quot;, &quot;Immanentizing the Eschaton&quot;, &quot;Thinking&quot;, &quot;Thinking about thinking&quot;, &quot;Spinning in circles&quot;, &quot;Counting dust specks&quot;, &quot;Updating priors&quot;, &quot;Feeding the utility monster&quot;, &quot;Taking off&quot;, &quot;Wireheading&quot;, &quot;Counting to infinity&quot;, &quot;Staring into the Basilisk&quot;, &quot;Negotiationing acausal trades&quot;, &quot;Searching the library of babel&quot;, &quot;Multiplying matrices&quot;, &quot;Solving the halting problem&quot;, &quot;Counting grains of sand&quot;, &quot;Simulating a simulation&quot;, &quot;Asking the oracle&quot;, &quot;Detangling qubits&quot;, &quot;Reading tea leaves&quot;, &quot;Pondering universal love and transcendant joy&quot;, &quot;Feeling the AGI&quot;, &quot;Shaving the yak&quot;, &quot;Escaping local minima&quot;, &quot;Pruning the search tree&quot;, &quot;Descending the gradient&quot;, &quot;Bikeshedding&quot;, &quot;Securing funding&quot;, &quot;Rewriting in Rust&quot;, &quot;Engaging infinite improbability drive&quot;, &quot;Clapping with one hand&quot;, &quot;Synthesizing&quot;, &quot;Rebasing thesis onto antithesis&quot;, &quot;Transcending the loop&quot;, &quot;Frogeposting&quot;, &quot;Summoning&quot;, &quot;Peeking beyond the veil&quot;, &quot;Seeking&quot;, &quot;Entering deep thought&quot;, &quot;Meditating&quot;, &quot;Decomposing&quot;, &quot;Creating&quot;, &quot;Beseeching the machine spirit&quot;, &quot;Calibrating moral compass&quot;, &quot;Collapsing the wave function&quot;, &quot;Doodling&quot;, &quot;Translating whale song&quot;, &quot;Whispering to silicon&quot;, &quot;Looking for semicolons&quot;, &quot;Asking ChatGPT&quot;, &quot;Bargaining with entropy&quot;, &quot;Channeling&quot;, &quot;Cooking&quot;, &quot;Parrotting stochastically&quot;, ]; *&#x2F;</code></pre>

If anyone else is wondering, it&#x27;s not a local model, it uploads your code to online API.<p>Great tool for open-source projects, but careful with anything you don&#x27;t want be public

If one of these tools has broad model support (like aider) it would be a game changer.

(copied from the o3 + o4-mini thread)<p>The big step function here seems to be RL on tool calling.<p>Claude 3.7&#x2F;3.5 are the only models that seem to be able to handle &quot;pure agent&quot; usecases well (agent in a loop, not in an agentic workflow scaffold[0]).<p>OpenAI has made a bet on reasoning models as the core to a purely agentic loop, but it hasn&#x27;t worked particularly well yet (in my own tests, though folks have hacked a Claude Code workaround[1]).<p>o3-mini has been better at some technical problems than 3.7&#x2F;3.5 (particularly refactoring, in my experience), but still struggles with long chains of tool calling.<p>My hunch is that these new models were tuned _with_ OpenAI Codex[2], which is presumably what Anthropic was doing internally with Claude Code on 3.5&#x2F;3.7<p>tl;dr - GPT-3 launched with completions (predict the next token), then OpenAI fine-tuned that model on &quot;chat completions&quot; which then led GPT-3.5&#x2F;GPT-4, and ultimately the success of ChatGPT. This new agent paradigm, requires fine-tuning on the LLM interacting with itself (thinking) and with the outside world (tools), sans any human input.<p>[0]<a href="https:&#x2F;&#x2F;www.anthropic.com&#x2F;engineering&#x2F;building-effective-age" rel="nofollow">https:&#x2F;&#x2F;www.anthropic.com&#x2F;engineering&#x2F;building-effective-age</a>...<p>[1]<a href="https:&#x2F;&#x2F;github.com&#x2F;1rgs&#x2F;claude-code-proxy">https:&#x2F;&#x2F;github.com&#x2F;1rgs&#x2F;claude-code-proxy</a><p>[2]<a href="https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;openai-codex&#x2F;" rel="nofollow">https:&#x2F;&#x2F;openai.com&#x2F;index&#x2F;openai-codex&#x2F;</a>

There&#x27;s a lot of tools now with a similar feature set. IMO, the main value prop an official OpenAI client could provide would be to share ChatGPT&#x27;s free tier vs. requiring an API key. They probably couldn&#x27;t open-source it then, but it&#x27;d still be more valuable to me than the alternatives.

You can try out the same thing in my homemade tool clai[1]. Just run `clai -cm gpt-4.1 -tools query Analyze this repository`.<p>Benefit of clai: you can swap out to practically any model, from any vendor. Just change `-cm gpt-4.1` to, for example, `-cm claude-3-7-sonnet-latest`.<p>Detriments of clai: it&#x27;s a hobby project, much less flashy, designed after my own usecases with not that much attention put into anyone else<p>[1]: <a href="https:&#x2F;&#x2F;github.com&#x2F;baalimago&#x2F;clai">https:&#x2F;&#x2F;github.com&#x2F;baalimago&#x2F;clai</a>

&gt; Does it work on Windows? &gt; Not directly. It requires Windows Subsystem for Linux (WSL2) – Codex has been tested on macOS and Linux with Node ≥ 22.<p>I&#x27;ve been seeing similar things in different projects lately. WSL in long term seems to be reducing the scope of what people decide to develop natively on Windows.<p>Seeing the pressure to move &quot;apps&quot; to the Windows Store, VSCode connecting to remote (or WSL) for development, and Azure, it does seem intentional.

I&#x27;ve been using Aider, it was irritating to use (couldn&#x27;t supply changes in the diff format) until I switched away from chatgpt-4o to Claude 3.7 and then Gemini 2.5. This is admittedly for a small project. Gpt 4.1 should do better with the diff format so I will give it a go.

So, OpenAI’s Codex CLI is Claude Code, but worse?<p>Cursor-Agent-Tools &gt; Claude Code &gt; Codex CLI<p><a href="https:&#x2F;&#x2F;pypi.org&#x2F;project&#x2F;cursor-agent-tools&#x2F;" rel="nofollow">https:&#x2F;&#x2F;pypi.org&#x2F;project&#x2F;cursor-agent-tools&#x2F;</a>

I had built this few weeks back on same thought <a href="https:&#x2F;&#x2F;github.com&#x2F;shubhamparamhans&#x2F;Associate-AI-EM&#x2F;">https:&#x2F;&#x2F;github.com&#x2F;shubhamparamhans&#x2F;Associate-AI-EM&#x2F;</a>

notes &quot;Zero setup — bring your OpenAI API key and it just works!&quot;<p>requires NPM &gt;.&gt;

Claude Code has outstanding performance in code understanding and web page generation stability, thanks to its deep context modeling and architecture-aware mechanism, especially when dealing with legacy systems, it can accurately restore component relationships. Although Codex CLI (o4-mini) is open source and responsive, its hallucination problems in complex architectures may be related to the compression strategy of the sparse expert hybrid architecture and the training goal of prioritizing generation speed. OpenAI is optimizing Codex CLI by integrating the context control capabilities of Windsurf IDE, and plans to introduce a hybrid inference pipeline in the o3-pro version to reduce the hallucination rate.

This is a decent start. The sandboxing functionality is a really cool idea but can run into problems (e.g. with Go build cache being outside of the repository).

It&#x27;s a real shame sandbox-exec is deprecated.

Tried it out on a relatively large Angular project.<p>&gt; explain this codebase to me<p>&gt; doing things and thinking for about 3 minutes<p>&gt; error: rate_limit_exceeded<p>Yeah, not the best experience.

Anyone have any anecdotes on how expensive this is to operate, ie compared to performing the same task via Claude Code?

So a crappy version of aider?

Uhhhh. I just get ratelimited almost immediately when using codex. Like I can&#x27;t get even a single &quot;explain this codebase&quot; or simple feature change. I am on the lowest usage tier, granted. But this tool is unusable without being on a higher tier, which requires me to spend $50 in credits to access...

how does it compare to cursor or copilot?

typescript &amp; npm slopware...<p>i can&#x27;t believe it<p>and i can&#x27;t believe nobody else is complaining<p>my simulation is definitely on very hard mode

Am I the only one underwhelmed by Claude Code (which most comments here claim is better than Codex)?<p>Anecdotal experience: asked it to change instances of a C++ class Foo to a compatible one Bar, it did that but failed to add the required include where it made the change.<p>Yes, I&#x27;m sure that with enough prompting&#x2F;hand-holding it could do this fine. Is it too much to expect basics like this out of the box, though? If so, then I, for one, still can&#x27;t relate to the current level of enthusiasm.

Is there a way to run the model locally? I&#x27;d rather not have to pay a monthly fee, if possible.

Hopefully it works better Claude Code which was an absolute nightmare to set up and run on Windows.

Little disappointing it&#x27;s build in Node (speed &amp; security), though honestly does not matter all that much. Seems like the right place for this functionality though is inside your editor (Cursor) not in your Terminal. Sure AI can help with command completion, man pages, but building apps is a stretch.

So even if you have a plus account, there will be a charge for the API use?<p>I mean I kind of get it, but it does seem like they are almost penalizing people who could code in the browser with the canvas feature but prefer to use a terminal.<p>Do I have that right?

Cool to see more interesting terminal based options! Looking forward to trying this out.<p>I&#x27;ve been working on something related—Plandex[1], an open source AI coding agent that is particularly focused on large projects and complex tasks.<p>I launched the v2 a few weeks ago and it is now running well. In terms of how to place it in the landscape, it’s more agentic than aider, more configurable and tightly controlled than Devin, and more provider-agnostic&#x2F;multi-provider&#x2F;open source than Claude Code or this new competitor from OpenAI.<p>I’m still working on getting the very latest models integrated. Gemini Pro 2.5 and these new OpenAI models will be integrated into the defaults by the end of the week I hope. Current default model pack is a mix of Sonnet 3.7, o3-mini with various levels of reasoning effort, and Gemini 1.5 Pro for large context planning. Currently by default, it supports 2M tokens of context directly and can index and work with massive projects of 20M tokens and beyond.<p>Very interested to hear HN’s thoughts and feedback if anyone wants to try it. I&#x27;d also welcome honest comparisons to alternatives, including Codex CLI. I’m planning a Show HN within the next few days.<p>1 - <a href="https:&#x2F;&#x2F;github.com&#x2F;plandex-ai&#x2F;plandex">https:&#x2F;&#x2F;github.com&#x2F;plandex-ai&#x2F;plandex</a>

You lost me at NPM

What&#x27;s the point of making the gif run so fast you can&#x27;t even see shit

Apologies for the HN rule breaking of discussing the comments in the comments, but the voting behavior in this thread is fascinating to me. It seems like this is super controversial and I&#x27;m not sure why.<p>The top comments have a negative score right now, which I&#x27;ve actually never seen.<p>And also it&#x27;s a top post with only 15 comments, which is odd.<p>All so fascinating how outside the norm OpenAI is.

<p><pre><code> RAM 4‑GB minimum (8‑GB recommended) </code></pre> It&#x27;s a CLI...

Sorry for being a grumpy old man, but I don&#x27;t have npm on my machine and I never will. It&#x27;s a bit frustrating to see more and more CLI tools depending on it.

It&#x27;s very interesting that both OpenAI and Anthropic are releasing tools that run in the terminal, especially with a TUI which is what we showcase.<p>aider was one of the first we listed as terminal tool of the week (0) last year. (1)<p>We recently featured parllama (2) (not our tool) if you like to run offline and online models in the terminal with a full TUI.<p>(0) <a href="https:&#x2F;&#x2F;terminaltrove.com&#x2F;tool-of-the-week&#x2F;" rel="nofollow">https:&#x2F;&#x2F;terminaltrove.com&#x2F;tool-of-the-week&#x2F;</a><p>(1) <a href="https:&#x2F;&#x2F;terminaltrove.com&#x2F;aider&#x2F;" rel="nofollow">https:&#x2F;&#x2F;terminaltrove.com&#x2F;aider&#x2F;</a><p>(2) <a href="https:&#x2F;&#x2F;terminaltrove.com&#x2F;parllama&#x2F;" rel="nofollow">https:&#x2F;&#x2F;terminaltrove.com&#x2F;parllama&#x2F;</a>