← sources

OpenAI keeps making its subscriptions worse, and somehow the answer is always "just pay more"

source: reddit (reddit:OpenAI)published 2026-11-01 15:37ingested 2026-11-02 02:39

retained text

# TL;DR: Over the past few months, OpenAI has brought back five-hour limits, progressively reduced and eventually removed the grace period after hitting usage limits, briefly introduced Luna Reserve before it disappeared from my account, and now plans to merge ChatGPT and Codex usage without telling us what the combined allowance will be. Plus users seem to be losing some of the biggest advantages they had over Claude, while the response to complaints is increasingly "just upgrade." I'm finding it harder and harder to justify paying for ChatGPT when I already prefer Claude for coding.

I've been using ChatGPT Plus for quite a while now, and I'm genuinely struggling to understand what OpenAI is doing with its subscriptions.

I already use Claude extensively for coding, and I personally find Opus 5.5 considerably better, especially for frontend work, UI design, and more complex projects. But I've kept ChatGPT Plus because it had one massive advantage: I could use ChatGPT for everyday conversations, research, brainstorming, and higher reasoning without worrying about consuming my Codex allowance.

That separation alone made keeping both subscriptions worthwhile. Unfortunately, it looks like that's going away too.

# How Codex went from generous to increasingly restrictive

Remember when OpenAI removed the five-hour limits? That was genuinely fantastic. You could sit down and work on a project without constantly worrying about when your session would end. It was one of the biggest advantages Codex had over Claude Code, and it made the subscription feel substantially more generous.

Then OpenAI brought those limits back for Plus users.

Disappointing, sure, but what bothers me even more is how they've progressively changed what happens when you actually hit your usage limit.

Codex used to let active sessions continue for quite a while after reaching the limit, sometimes for hours. People understandably started abusing that to squeeze out more usage, so I don't particularly blame OpenAI for wanting to address it.

But then they shortened the continuation period, shortened it again, and eventually removed it entirely.

Now, in my experience, **the moment you exhaust your allowance, the agent just stops.** It doesn't matter if it's halfway through making changes to your codebase. There's no opportunity to wrap up the task, clean up unfinished changes, or even leave the project in a consistent state. You can end up with partially modified or broken code simply because the agent happened to hit its limit at the wrong moment.

I find that pretty ridiculous for a tool that's supposed to be used for actual development.

I'm not asking OpenAI to give everyone another three hours of free compute. I understand why the original continuation system wasn't sustainable. But surely there's a reasonable middle ground between allowing people to abuse the grace period and abruptly killing an agent halfway through its work?

They actually seemed to have a sensible solution with Luna Reserve, which introduced a separate fallback allowance using a cheaper model once your main quota was exhausted. I had access to it for a few days, thought it was a great idea, and then it silently disappeared from my account. Haven't seen it since.

So we've gone from having no five-hour limits and generous continuation, to having those limits reinstated and agents abruptly stopping when they're reached. A potentially useful fallback appeared along the way, but that disappeared too.

I don't see how that's an improvement for the user.

# And now they're merging ChatGPT and Codex usage

Starting in December, ChatGPT and Codex will share the same token-based allowance, with Instant chat being the exception.

This is probably the most frustrating change for me personally because it removes one of the main reasons I chose ChatGPT over Claude in the first place.

I regularly use ChatGPT for long conversations and research, sometimes spending hours discussing something completely unrelated to development. I could do that without affecting my ability to use Codex later. It was a genuinely useful distinction between ChatGPT and Claude, where chat and coding already share an allowance.

With the new system, I'll have to think about how much of my coding allowance I'm consuming every time I have a substantial conversation using a higher reasoning model.

Maybe OpenAI will increase the combined allowance enough to make this a nonissue. I'd love that. But the announcement doesn't actually give us any numbers.

We have no idea whether the combined allowance will preserve the practical capacity of the two previously separate limits, let alone improve it.

Instead, we're told that most users shouldn't need to wait for their usage to reset more often than they do today.

That's reassuring for casual users, I guess, but what about people who actually use both ChatGPT and Codex heavily? Those are precisely the users who benefited most from having separate allowances, and we're being given almost nothing concrete about how the change will affect us.

And while Plus users seem to be getting hit particularly hard, the higher subscription tiers haven't exactly had a smooth ride either.

The whole Pro pricing situation has been a circus. OpenAI restructured the plans at considerably worse usage-per-dollar ratios, then announced doubled allowances roughly a month later, largely restoring the value offered before the restructuring. Existing $200 subscribers were even facing a reduction that ultimately got canceled.

I'm glad they reversed course, but presenting the restoration of previously available value as some massive improvement is a little ridiculous. Meanwhile, Plus users haven't received any comparable increase despite the changes coming to their usage.

# Since when did $20 become pocket change?

One thing I find particularly strange about these discussions is how quickly people dismiss complaints about usage limits.

"You're only paying $20, what do you expect?"

Since when did $20 a month become an insignificant amount of money? And why does every conversation about AI limits seem to end with someone suggesting that you should just upgrade to the $100 plan?

I think some people genuinely forget that not everyone lives in the US or earns a six-figure salary. There are countries where **$100 is literally what many people earn in an entire month.** Even in wealthier countries, $1,200 a year for a single software subscription is a pretty significant expense.

There are students, hobbyists, independent developers, people learning to code, and ordinary users who just want to make good use of these tools. Not everyone is running a business or generating revenue from AI.

I don't think it's reasonable to treat a $100 subscription as the obvious solution whenever someone complains about the usability of a $20 product, especially when much of the functionality they're complaining about was previously available to them.

And I'm not arguing that a $20 subscription should provide unlimited access to the most expensive models. The original continuation system was clearly being exploited, and I completely understand that companies need to establish sustainable limits.

But these complaints aren't simply about wanting more tokens. They're about a product that keeps becoming more restrictive, features disappearing, and usage policies changing without enough transparency about what customers are actually getting.

There's a difference between accepting reasonable limits and being expected to celebrate every reduction in what your subscription provides.

# The whole "you're being subsidized" argument

I've also seen people calculate their token consumption using API pricing, compare that figure to their subscription fee, and conclude that OpenAI must be losing hundreds or thousands of dollars on them.

Except API prices aren't the actual cost of inference. They're commercial prices with margins built in. Using $500 worth of API-priced tokens doesn't mean OpenAI spent $500 serving those requests.

I'm sure some particularly heavy users genuinely cost more to serve than they pay. There are also plenty of expenses involved in operating these services beyond inference itself. But without knowing the actual costs and economics of the subscriptions, I don't see how anyone can confidently calculate how much they're supposedly being subsidized just by looking at API prices.

**And even if OpenAI really is subsidizing these subscriptions heavily, how does that make customers' complaints invalid?**

OpenAI chose to offer these products at these prices. People subscribed based on what they were getting. If the economics require changes, that's understandable, but it doesn't mean customers have to be happy about receiving less for the same money.

I'm perfectly fine with reasonable limits. What I do expect is some consistency and transparency about what I'm paying for. Having coding agents abruptly abandon unfinished work, removing useful fallback features, and merging major usage allowances without publishing the actual numbers makes it increasingly difficult to know what kind of experience I'm paying for from one month to the next.

# Why I'm seriously considering switching

All of this is happening while the competition is getting increasingly attractive.

I already prefer Claude for development, and while I haven't extensively tested it for everyday conversations, I'm finding fewer reasons to keep ChatGPT Plus alongside it. Anthropic has its own problems with usage limits, of course, but separate and generous chat usage was the main reason I could justify maintaining both subscriptions.

I don't particularly care about image generation, and connectors are easy enough to set up elsewhere. There are certainly things ChatGPT does well, but I struggle to identify a remaining advantage that matters enough to justify paying for it over Claude.

The funny thing is that my biggest reason for staying has very little to do with the product's current capabilities. It's the years of conversations, accumulated memories, personal preferences, and ongoing context I've built up on ChatGPT. Migrating all of that would be an absolute pain.

I never thought I'd reach a point where the inconvenience of leaving was a stronger reason to keep my subscription than the actual benefits of using it.

I'm curious how other long-time subscribers feel, I've already cancelled my subscription. 

document fetched 2026-11-02 00:27 · canonical https://reddit.com/r/OpenAI/comments/1xl6431/openai_keeps_making_its_subscriptions_worse_and