OpenAI, please reconsider merging ChatGPT and Codex usage limits. You’re giving people like me a reason to switch to Anthropic.
I’ve been using ChatGPT and Codex together for serious engineering and software development work, and the recent announcement about merging their usage limits is genuinely disappointing.
I think OpenAI might be underestimating why some of us chose its ecosystem over Anthropic in the first place.
For me, one of the biggest advantages of ChatGPT has been the ability to do most of the thinking and engineering work before touching the actual code.
I can spend hours discussing system architecture, analyzing algorithms, working through difficult technical problems, reviewing design decisions, and figuring out exactly what needs to be built. Then, when the plan is solid, I move to Codex for implementation and testing.
That separation is incredibly useful.
I’m not wasting coding capacity on discussions, and I’m not rushing important engineering decisions because I’m worried about exhausting the same allowance I need to write and test code.
That’s a real advantage OpenAI has had over Anthropic, at least for the way I work.
And now it sounds like you’re planning to remove it.
I understand that Instant chat will remain unlimited. But let’s be realistic: Instant isn’t a replacement for High or Extra High reasoning when you’re working on complex engineering problems.
Sometimes a difficult architectural decision takes several rounds of deep reasoning. Sometimes you need to challenge an approach, explore alternatives, and go back to the drawing board before writing a single line of code.
If all of that starts eating into my Codex allowance, I’m effectively being penalized for doing the engineering work properly.
And here’s the part I really think OpenAI should consider.
Why would I stay with OpenAI instead of moving to Anthropic?
Claude and Claude Code are already very capable. In several coding and agentic engineering benchmarks I’ve followed, Anthropic’s models have performed exceptionally well, sometimes outperforming OpenAI’s alternatives.
I’ve continued using OpenAI partly because the overall workflow made more sense for me. Even when another model looked stronger on a particular benchmark, the ability to separate deep engineering discussions from coding was a meaningful advantage.
But if both ecosystems are going to operate with shared usage limits, that advantage becomes much harder to justify.
At that point, I’ll simply compare which provider gives me better engineering results for the money I’m paying.
And frankly, Anthropic becomes a much more attractive option.
I don’t think I’m alone in this.
People building complex software, running coding agents, or working on long-term engineering projects aren’t just counting how many chat messages they can send. We’re looking at how much actual work we can get done with a subscription.
Combining two products into one convenient interface is a good idea. I have no objection to that.
But combining their interfaces doesn’t mean you have to combine their usage allowances.
Those are two completely different decisions.
If OpenAI needs to manage compute costs, I understand. I’m not asking for unlimited access to expensive models.
What I would like to see is a reasonable solution that preserves the existing workflow. Keep the advanced ChatGPT reasoning allowance separate from Codex, reserve a guaranteed amount of reasoning capacity, or at least let subscribers choose between separate and shared limits.
And before making this change, please publish the actual numbers. How much usage will we get? How will High and Extra High reasoning affect Codex capacity? Will someone who regularly uses both products be able to complete the same amount of work as before?
Without those details, it’s difficult to see this as an improvement.
I’ve invested a lot of time in building my workflow around OpenAI’s tools. I would prefer to keep using them.
But I’m not going to keep paying for an ecosystem simply out of habit if one of the main reasons I chose it disappears.
If this change makes advanced ChatGPT reasoning compete directly with my Codex usage, I will seriously consider moving my engineering work to Claude and Claude Code. That decision could come down to this change alone.
And I suspect quite a few other developers and engineers will be asking themselves the same question.
OpenAI has an opportunity to make ChatGPT and Codex work better together.
I just hope that doesn’t come at the cost of what made the combination valuable in the first place.