Ask HN: Claude Code or Codex?

[czeizel]

And why? I'd genuinely like to understand who is on which side and why. I'd also like to know, if you use one of them, what kind of subscription plan you have.

(For example, I use Claude (max) [...])

[kingkongjaffa]

I have personally been a heavy claude code user since it launched, I have a claude subscription at home and my employer pays for a claude account at work.

Until the launch of Fable I'd been very happy with claude code besides the outages where it could be offline for hours.

When Fable released it was a good step up from using Opus.

When it was taken away and then reinstated under some weird billing and vague notion of how it would be available in the long run, I noticed the model itself was worse than the brief time it was up before the US gov shut it down.

Now I'm evaluating Codex at home for personal use. I may switch to using chatGPT at home and whatever work provides (which is currently Claude).

Codex and 5.6 Sol seem just as good as Opus and Fable right now, but I only have 1-2 weeks testing.

Keeping the flagship models in the basic subscription is the main driver for me evaluating chatGPT again, I hadn't really used it since about GPT4 (I missed all of the O-models like o1, o3) and I moved to Claude when it was clearly better.

The ergonomics of claude skills are better than codex's agent skills imo.

ChatGPT and codex have more generous usage limits it seems.

[OleksandrC]

One significant advantage of Codex subscription over Claude is that you can use any agent harness with it. Meanwhile Anthropic insists that Claude subscription must only be used with Claude Code itself - and I have significant gripes with the quality of Claude Code in general.

Also, Fable access was now removed from small 20$ Claude plan, while Codex includes access to comparable model (gpt-5.6-sol) even on small subscription.

[theowaway213456]

You can also build your own harness for Claude, right? I think that's not against their ToS?

Not saying it's ideal, just saying it's an option

[deflator]

I switched to Codex in November 2025 because I couldn't stand the smarmy, "You're absolutely right" anymore. Codex at the time offered less sycophancy and more pushback when I was actually wrong. I also feel it writes cleaner code since it follows my instructions better to not write unrequested code, not create unnecessary abstractions, and not reinvent any wheels. I use LLMs on a tight-leash in HITL for web development, monitoring traces as it goes. I rarely use it for more than just writing code that I specify and I rarely let it "have it's head" and work unmonitored. I use it on the API plan since I need it to be reliable. No outage issues that I can recall.

[taf2]

August 2025 for me but now days I’m much more extreme- it does it all max sol in mornings and then I do other things in the afternoon. I get more done than before with far less stress.

[sergiotapia]

Sincerely try Cursor and Grok 4.6. I think I can't go back to a TUI anymore for regular dev work. We used to stomach the CLI because it had the better models, but that's not the reality today.

[aykutulis]

I use Claude Max 5x

Although Codex and Claude seem to have the same price for similar plans, Claude's tokens run out much faster. But I can't complain about that because I get much better results. Especially when working with very complex systems, I don't need to repeatedly revise Claude's changes as much as Codex, so in some cases it might even be cheaper.

Also, in my experience, Claude tends to follow rules and instructions more consistently

[217]

https://github.com/can1357/oh-my-pi

https://stencil.so/blog/prewalk

most advanced, configurable and capable harness of all time, and i use both 200$ subs with it. Has a huge learning curve though so if you're not into that i'd go with whatever you already use the most

[czeizel]

hadn't come across either of these before, thanks. using both behind the same harness sounds really interesting.

do you still choose claude vs codex depending on the task, or has the harness made them mostly interchangeable? what took the longest to learn?

[217]

It's vibes mostly. Like, fable is better on frontend while 5.6sol is better at long running tasks with a lot of subtasks, harness did made them mostly interchangeable as well as able to have them both participate in solving a hard issue. Longest to learn is the ui (I'm more of a GUI guy but this is just too good) and all hundreds of features where you want to do something and then discover the harness already supports it out of the box but you never knew it, yet

[dimonb19a]

Each model has a different personality. There is some real layer underneath but it also depends on how you use it. For me GPT is a nerd and Claude i consider like a senior dev with deep knowledge. I use GPT only for audits and serious code logic because it can catch gaps and inconsistencies that Claude misses. But it always depends on your mindset, i think what really matters is a loop - you plus AI.

[beecasthurlbow]

I’ve been using both, but I find 5.6 sol is faster in terms of wallclock time and far less chatty (on my small projects in non-legacy code bases).

It “feels” like Sol gets off track more easily than Fable (it’s hard for Sol to ignore a stray remark). My theory is this comes from smaller context window - Codex uses a 350K token context window (that you can’t change), even though model can handle more…

[jdw64]

I use both. Claude I use up to 5, and Codex I use up to 20 for PRO.

Personally, I think Codex is better. Aside from the frequent resets, since the Fable incident, there have been cases where connections drop. For writing code, Codex generally does a better job. That said, I feel Fable is better for overall structure, but ultimately, it varies from person to person. I use both as my coding assistants

[wek]

I have Max subscription to both Claude Code and Codex. I use both fulltime throughout the day. I have them check each other's work, and route tasks they are best at to the right one.

The checking each other's work is so helpful.

New features, mockups, I use Claude Code. Deep fact-checking, I use Codex.

My harness matters at least as much as the model. I use the best harness and then invest a ton in it

[kkjh0723]

I use Claude Max, and I like that Fable can write code and run experiments on its own. I just check in occasionally to see how things are going and suggest the next direction. I expected Claude to be the clear favorite here, but there are more Codex users than I thought, so now I’m tempted to give it a try. That said, GPT translated this comment for me.

[czeizel]

the ability to let fable work independently, run experiments and only check in occasionally is probably claude's strongest argument here. i'm also surprised by how codex-heavy the replies have been, although the pricing and usage limits seem to explain part of it.

what kind of tasks are you comfortable leaving fable alone with, and how often does it take the project in a direction you need to undo? also, gpt translating the pro-claude comment is a nice ending :)

[bicepjai]

My high level methodology: Claude code fable for planning and execution. Codex sol for review. My constraint for one task convergence is codex saying no issues found on 3 consecutive run for the current diff along with code coverage with tests (95% with unit tests and 85% for integration tests)

[palguna26]

I personally use Codex(plus) a lot, with sol+ponytail, u get concise to the point answers rather than long explanation that openai models are known for, and till date im really satisfied with its coding performance. I also use opencode to try out new openweight models as well, currently im testing out k3.

[w4v3z]

Codex for raw autonomy/speed and near-zero added cost if you're already paying for ChatGPT; Claude Code if you want tighter interactive control and don't mind working within Pro/Max quota.

[alex7o]

Somebody had shared https://harnessrank.net/ but in reality I use Claude code as I am afraid to be banned

[Talpur1]

I mostly use claude, its actually good at context specially opus 4.8

[no-name-here]

> For example, I use Claude (max) [...]

What does […] mean in the OP post body? That you use other solutions than the 1 mentioned but are omitting them?

[czeizel]

no, sorry for the confusion. i meant [...] as a placeholder for the rest of the answer, as in: "i use claude (max) because of x, y and z..." not that i'm omitting other solutions.

- OP

[gokuljs]

I use Cursor and am happy with Composer :)

[eckelj]

I use claude and opencode for other models like qwen or deepseek. love both.

[invincible703]

i have used both and as of today i would say codex is better value for money and overall use cases. It is on par with the claude code and arguably better in some cases and costs way less that claude

[Squeeze2664]

I heavily use the Pi harness with DeepSeek-v4-pro directly from the DeepSeek API almost exclusively and I have have barely ever felt the need for something different. I have some OpenRouter credits and API key ready to go if DeepSeek fails me, and that almost never happens. It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks. I use it for coding, code review, managing my systems in general, and even regular-old-questions. Pi is minimal in many regards including system prompt, tools, and UI, and I found this quality much more preferable than what the other harnesses provide. Currently I use only two extensions to extend Pi's capabilities: a web search tool and a page fetch tool. I might look into getting more as the need arises.

I used to be a heavy Claude Code user but I ended my subscription around the time this Fable nonsense started, as I really don't like the direction Anthropic is taking. I now spend less, get more done, have a better time doing it, and feel better about it.

[mathew_woo]

Honestly I haven't tried Codex yet, but from what I've used so far, Claude Code feels pricey but worth it.

[Grapeeee]

of course codex, its more humanlike

[czeizel]

what makes it feel more humanlike to you - the way it reasons through problems, the tone of the responses, or is it simply more comfortable to work with day to day?

[dimonb19a]

For me it is opposite

[czeizel]

interesting, that probably shows how subjective "humanlike" is here! thx for your reply!

[eadesola]

claude code all the way