Areas where Codex is better than ClaudeCode

I have been paying for and using ClaudeCode for a long time, and since Codex offered a one-month free trial, I decided to give it a try and have been using it for a little over a week now.

Compared to ClaudeCode, where I have done and tried many things due to the length of time I have used it, I have only just touched the surface with Codex.

Therefore, I haven't used it enough to make comparisons like 'ClaudeCode is better for X, and Codex is better for Y'.

Mostly Codex

However, just during this Golden Week, I have used almost nothing but Codex.

Mainly in Obsidian, but through the plugin Claudian, the basic operations and usage are generally the same whether it is ClaudeCode or Codex.

Of course, as I posted before, a big part of this is that I have been able to configure Codex to be similar to my ClaudeCode settings.

However, even in this short period, I discovered something that is overwhelmingly different.

Not worrying about token consumption

The biggest thing I felt with Codex is that I hardly have to worry about the token consumption limits that I was constantly concerned about with ClaudeCode.

This is mentally much easier than I thought it would be.

It is not that I am not paying attention at all; I still check the context percentage and token consumption frequently.

Even so, compared to ClaudeCode, I feel that token consumption is less than half, or rather, the rate of consumption is slower.

Even when I generate images using ChatGPT Images 2.0, as I posted yesterday, and try various things with random prompts, I don't hit any limits at all; in fact, I'm surprised and think, 'Is the limit not coming yet?'

It makes me feel like maybe I should just make Codex my main tool.

Where ClaudeCode is better

This is also an impression based mainly on Obsidian, but when I ask or instruct Codex to do something, the response is slow and the answers are long.

In contrast, ClaudeCode's answers are fast and precise.

It is true that a lot of information is accumulated in ClaudeCode's memory in Obsidian, but what I thought while using Codex was that it feels like it is thinking and thinking, and then it returns a long response, which takes time.

It is partly a matter of getting used to it, but ClaudeCode is the one that works the way I expect.

The Codex app is also good.

Although I mainly use Codex within Obsidian, I also use it in the Zed editor, just like ClaudeCode.

However, when I opened and used the same project in the Codex desktop app, the response was good, and the fact that I can use ChatGPT Images 2.0 normally is very nice.

Incidentally, the accuracy of the built-in voice input is good. I haven't bought a microphone yet, so I haven't used it that much, but for now, it's stable enough that I feel like the voice input in the Codex app is good enough.

It's also wonderful that if the Codex app is running in the background, it can be used in other apps on the machine or in browsers, not just in Codex.

I still need to use it more.

I started using Codex with a one-month free trial.

I'm not programming or building apps, so I don't know how it performs in those areas.

For the first two days, it was still GPT-5.4, so I wouldn't say it was bad, but I felt ClaudeCode was much better. However, since it updated to GPT-5.5, while the responses are long, it has become excellent in terms of Japanese interaction, and it's becoming fine for normal use as well.

After that, if I use it a bit more, align my habits with Codex's features, and do some fine-tuning and adjustments regarding settings like AGENTS.md, skills, and plugins, it seems like it could be used without any inferiority.

Within Obsidian, there are no problems at all.

Compared to my initial feeling of just trying it out, it's certainly good enough that even with my shallow experience so far, I feel like I can use it normally and that I could probably get by with just Codex.