Quick conclusion
I don’t pick a winner. I use ChatGPT, Claude, and Codex on the same blog, and I switch between them by job.
Right now Claude Code is the one doing most of the writing. ChatGPT handles brainstorming and my featured images. Codex sits in the background as a backup and as my coding tool. If you only want one to start with, I’ll tell you which one near the end, and it is not the one I use most.
Here is how the split actually looks in my day-to-day.
The three tools, in one line each
This is not a spec sheet comparison. It is how each tool earns its place in my workflow.
- Claude Code (desktop app): my main writer. The English it produces sounds the closest to a real person.
- ChatGPT (with Images 2.0): my thinking partner and image studio. The image quality is the best of the three by a clear margin.
- OpenAI Codex (desktop app): my backup writer and my coding tool. Slower to burn through usage, and in my experience its limits feel separate from ChatGPT.
I run all of these as desktop apps, and I lean on Claude’s Opus model for anything where the wording matters.
My actual pipeline
A finished post moves through four stages: research, drafting, fact-check and voice, then export and publishing. Each stage has a tool I reach for first.
| Stage | Tool I reach for | Why |
|---|---|---|
| Brainstorming and outline | ChatGPT | Fast back-and-forth when an idea is still rough |
| Drafting and voice | Claude Code | The most natural-sounding English |
| Structure and how-to steps | Claude Code (used to be Codex) | Lately Claude handles this too |
| Featured image | ChatGPT Images 2.0 | Best image quality of the three |
| Building support tools (not the post) | Codex | Cheap, steady coding |
A year ago my mental model was simpler: Claude for voice, Codex for structure and steps, ChatGPT for brainstorming and images. That is still mostly true. The shift is that Claude now does the structural work too, and Codex has moved off the writing pipeline almost entirely. These days I point Codex at coding jobs instead, like a small tool that downloads my analytics data. For that kind of work it is the most cost-effective of the three.
One detail matters more than people expect. Claude Code and Codex can both reach into my Obsidian vault and edit files directly through MCP. ChatGPT cannot. That single gap shapes a lot of my choices, and I come back to it below.
Where each one wins (and loses)
I pay for all three, so this is not theory. Here is the honest version.
ChatGPT
Where it wins: images. ChatGPT Images 2.0 is the best image generator I use, and it is not close. My featured images come from here.
Where it loses: it cannot touch my vault. There is no MCP connection, so it cannot read or edit my Obsidian notes the way my other tools do. Honestly, if that gap were closed, I might not have needed to upgrade Claude at all.
Claude
Where it wins: the writing. Claude’s English is the closest to how a real person talks, and the Opus model is steady in a way I have come to rely on. The common advice is that a smaller model is fine for blogging. That may be true, but once I got used to Opus I did not want to go back.
Where it loses: cost. Claude burns through usage fast, and it feels expensive next to the others. I still think it is worth paying for, but I notice the meter.
Codex
Where it wins: coding, and economy. It uses up its allowance more slowly than Claude, and in my experience its limits feel separate from ChatGPT, so one subscription effectively buys me two lanes. For building small tools it is my first pick.
Where it loses: images. Its image generation is underwhelming. It does not come close to what ChatGPT Images 2.0 gives me.
The week I hit the wall
For a long time I told myself that a Pro plan was plenty for someone who just updates a blog.
Then, mid-week, in the middle of writing and exporting a post, I hit Claude Pro’s weekly usage limit. The plan was capped for the rest of the week. That left me roughly four days with almost nothing, and there was no way to work around it.
So I upgraded. Claude Pro is $20 a month, or about $17 a month if you pay annually, which is what I had been on. The next step up is the Max plan at $100 a month for the 5x tier (there is also a 20x tier at $200). Max is monthly only, with no annual discount.
Here is the part I did not expect. When I upgraded, the unused portion of my annual Pro plan was refunded as credit, and that credit was larger than the new charge.
The Max charge was $100. The refund for my unused Pro time was $157.95. After the math, I owed nothing, and I had $63.75 in credit carried forward. I effectively moved up a tier and came out ahead.
I am treating Max as a one-month trial for now. One small wish: the jump from Pro to Max 5x is a big one, and 5x is honestly more than I need. A 2x or 3x tier in between would fit me perfectly.
How I decide per article
My default is Claude. If a post lives or dies on its voice, Claude writes it.
When my Claude usage is running low, I move the work to Codex. It is the natural backup, and it keeps the pipeline moving without burning my Claude allowance. The article type plays a role too: a cornerstone piece where the writing carries everything goes to Claude, while more structured, step-by-step posts are fine on Codex.
There is no strict rule here. I watch the usage meter, look at what the post needs, and pick.
If you only pick one
If you want the most consistent quality, pick Claude. Just know it is the most expensive of the three, and the usage adds up.
But if I had to recommend a single starting point for most people, I would say Codex. The reason is practical: it pairs with ChatGPT, and in my experience the two do not fight for the same usage, so one path gives you a capable writer and the best image tool. Claude is steadier, but you pay for that steadiness.
In short: best quality, Claude. Best value and easiest on-ramp, Codex plus ChatGPT.
FAQ
Is Claude Max worth it just for blogging?
For me, yes, but with a caveat. I only upgraded after Pro’s weekly limit blocked me mid-post. If you regularly run out on Pro, Max is worth it. If you don’t, Pro is plenty.
Can ChatGPT edit my Obsidian vault?
No. ChatGPT has no MCP connection to my files, so it cannot read or edit my notes directly. Claude Code and Codex can. That is a real reason I lean on them for the actual writing.
Which one is cheapest to start with?
ChatGPT Plus and Claude Pro are both $20 a month. Codex is the most economical in practice because it burns usage slowly, and in my experience its limits are separate from ChatGPT’s.
Do I really need all three?
No. I use three because each one is clearly best at something for my workflow. If you are starting out, one tool is fine. Add another only when you hit a specific wall, the way I did.
Next steps
If you want the bigger picture of how this fits together, here are two related posts:
- How I Run 5 Blogs Solo Using AI and a Second Brain
- How I Write an English Blog as a Non-Native Speaker (with AI)
I’ll keep updating this as my setup changes. The Max trial ends in a month, and I am curious whether I stay.