OpenAI restores 5-hour Codex and Work limits for ChatGPT Plus users
9to5mac.com
9to5mac.com
Such a casino vibe.
If you see any sort of timer anywhere, you are in a Skinner's box simulator. You are being subjected to positive and negative reinforcement via the reward schedule. This is habit forming.
it's like nitrous oxide, you're sure to have that epiphany with just one more pull
The timer only starts when we message a model, and we always want the timer to be counting down. So if a timer resets at 3 AM, we are incentivized wake up at 3 AM and start something just to get the timer running. Mercifully, unlike actual video games, automating this isn't cheating.
The existence of timer resets also incentivizes us to spend our usage as quickly as possible. No way to know when Tibo's gonna push his reset button. Ideally, we want to be at zero percent remaining when he does. Any usage left before reset is usage that was left on the table. The new 5h limits will make this much more difficult, and this is probably by design.
Heh, it's just like my mobile gaming days... Deal with this long enough and eventually you can see the game mechanics like Neo sees the Matrix. Never thought I'd fall into this rabbit hole again.
I disagree about the "what did they build" angle though. These subscriptions are providing me with truly absurd amounts of value. Gacha games deserve to be mercilessly botted to death, but AI is world changing technology.
The end result is that I’m considerably under quota every week, so I’m sure margins are comfortable for OpenAI. But it means my work doesn’t get interrupted, and I can use heavier models and fast mode liberally, because I’m not going to hit quota anyway, which means my sessions are shorter and more productive.
It’s hard to get anything “done done” when you get a quota limit, just as you’re getting into a good flow.
It's the opposite for me. The more money I pay, the more pressured I feel to squeeze out every last percent of usage out of the subscription. I can only know peace when the meter is at 0% remaining.
God help me if the OpenAI guys announce an incoming reset. If it's not going fast enough I'll just find new things for the AI to do, usually a massive parallel code review. They can always reset without any warning too, they did it after breaking the news this thread is about. Slept with an empty tank and woke up with a full tank. Wish my car worked this way.
At least I'm actually doing things instead of procrastinating.
I've been running statistics on my local codex transcripts in order to try and figure out how many API credits are included in the subscription. So far my results are hovering around 3k credits for Plus and 15k credits for Pro 5x, which is aligned with what they advertise.
> Such a casino vibe.
No no no! It means we're all in it together! Feel the parasocial vibes!
Free drinks? Keep customers happy? Reward your heaviest users? Casino vibes
It could leave them out of availability later when they need it.
It's a slightly entitled complaint but when you 'waste' 80% of your remaining usage because you get a reset you weren't expecting, there's a ping of disappointment there.
I'll be switching back.
To an extent I get it, this is not like a normal SaaS where you just buy seats, but their commercial offerings, especially B2C ones, feel like complete amateur hour. If they were selling literally anything else I'd have ran for the hills a long time ago.
Seems fair, but then maybe the 5h usage should also drain slower during off-peak hours? Perhaps it already does.
> and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience.
Wouldn't this one be fixed by making the 5h limit a guardrail you can opt out of? If so, then this doesn't work as justification for a mandatory limit.
Nothing shows that it is other than normal and healthy behavior that is produced from within a rational mental state. The complaint about scheduling is also normal and rational.
The closest thing to "pathological" that I'm seeing here in these comments is trendy misuse of the term.
which TZ?
> There are no off-peak hours.
Probably US west coast hours, but there's no public stats
At this point the 40 USD/120 USD Cursor Team subscription suddenly becomes a lot more attractive. You'll get more generous limits and a single monthly limit.
The rest is garbage but I guess is how they make their money.
more than once a quarter? nah. would also delete it if I could
I'd rather the site asked me to pay directly, or not at all. I will not see any ads I can avoid.
The annoying things is that you gave bigger tasks and is such session finished maybe 70% and after 5h pool resets cache will expire so that if you "please continue" it will wipe out again a lot of your 5h pool)
I could burn through my Opus usage in 1 hour whereas with Sol, I can go 3–4 days.
* OpenAI was ~$170 per week in value
* Claude was around $500 per week in value.
* Ollama Cloud is giving me about ~$160 per week in value.
So right now (thing constantly change in the AI world), its not more generous then Claude.
Geofencing has been quite good at creating feeling of being ripped off in the developed countries - from which the bulk of the companies revenue come from - that companies seems to be dropping it lately.
Anyway - China's advances seems to have caught the big 2 unprepared, so price war seems to be going on.
And there are Chinese resellers for frontier models if you are desperate for high end frontier tokens.
Tighter limits, forcing people to higher plans. Most will go as they are addicted and dependent on those tools. They need to use to finish the project they started as no human will touch that pile of code. If you didn't saw this coming, as it has been the standard business strategy from all kinds of SaaS in the last years, you haven't been paying attention.
It’s more expensive, yes, but I assume most people here use it for something that is worth spending money on?
I’m basically replicating myself multiple times for a ~10% salary surcharge.
In my mind it’s impossible to not see it that way, unless you just want to capture the subsidy surplus for free.
Yes, LLMs are awesome and let me do things faster but I would be using them far less if my only option was the API. I used coding agents prior to subscriptions (Aider) and spent <$200 total before abandoning it due to cost. The results were good, but not worth the price for me.
I'm happy to answer more questions if that doesn't cover what you were looking for. I'm not sure if "coding" is all you wanted or if you wanted more details.
What would you say is the split of your token use (between main job / side project)? Do you use agents for everything?
I'm asking because I feel I'm weird in my usage, which is very much structured like 1. organising the context and then 2. send off requests to 1-2 models and continue down the rabbit hole from there. I've found agents take longer and make me learn/retain less about the system I'm building.
It fluctuates, but never less than 30% on either and I get to 95%+ of my weekly usage. I also spend some tokens on “personal” stuff, like with my Obsidian notes or helping to manage other tech in my life (Home Assistant, Unifi, NAS, 3D Printer, other servers).
> Do you use agents for everything?
Increasingly so. It’s just so much faster in many cases or even if it takes the same time or more, I can do something else while it’s working. All my communications are 100% me, (sms/iMessage/HN/Slack/etc) but anything I consider “busy work” I try to farm out. I find the edges of what it can do then back off a bit or stop using it for that task.
> I'm asking because I feel I'm weird in my usage…
I like using skills to replicate the steps I want taken. It’s far from perfect and I want a way to enforce stricter workflows but skills with “Step 1, Step 2, Step 3…” do work. Currently once I’ve filled out a ticket (often using an agent to bounce ideas or grill me on my ideas) I can hand that off to an agent to plan, it will do its planning, present the plan, then spin up a worktree running an isolated dev stack, do the work (asking me questions as needed), then tell me it’s done (providing me with one-click login urls for the various roles in my system) with a link to the running dev stack, I then review the work, it kicks off 2 reviews, and fixes any issues they surface. Then a PR is created, this is normally when I look at the code if I need to, then 2 agents review the code (more of a smoke test but they find things), the original agent waits for the agents on GH Actions to finish, then deals with their findings, then it kicks off another review (up to 2 rounds). I decide then if any remaining issues are worth not merging (if you’ve used LLMs for code review you know they can go on endlessly with increasingly unreasonable edge cases).
All that said, my workflow is flux as I try out new things regularly. Does that slow me down? Perhaps, but things are changing too quickly to “settle in” to any workflow just yet. You could compare it to the JS Framework Cambrian explosion but it’s easier to switch.
> I've found agents take longer and make me learn/retain less about the system I'm building.
Fair point, some parts I scrutinize, and some areas I’m less concerned. I have E2E tests on the important flows, I perform manual testing constantly, there are extensive unit tests (front and backend), and the code is structured and abstracted better for testing than it’s ever been. All that to say is I feel confident of foundation, some of which was built pre-LLM and some of it was built by LLMs.
I don’t mind the 5 hour window but this is a reminder to not hinge your workflows on any one tool.
Maybe if we have some breakthrough in how to get equivalent ai capacity out of less compute it’ll seem crazy in retrospect.
But the magnitude of flops per token on SOTA models is mind boggling huge.
However I think there is still a significant runway for these models to scale, so there will always be some sort of offering from providers. I can't imagine that our current use of the context window will be how that looks in a handful of years.
Any other outcome would be pretty terrible.
There's your answer
I don’t think everyone here is using Claude/GPT/… to create indie games?
If you’re using it for work then waiting another week/day/hour is a business expense, effectively.
Maybe it's like this guy I know who purchased a new-ish Bugatti but all the service is done by a local mechanic, during his time off, in a barn.
At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness.
I tried for example Deepseek V4 Pro for a bit (via API pricing), but it's _more expensive_ than Sol via subscription, even though $1.7/Mtok is a lot less than 20$/Mtok.
> At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness.
Could you elaborate on that?
Off topic: You did the ISBN visualization thing! Awesome!
More. On CC I'm currently burning ~$5k a day (according to claudes own metrics) and I use up the weekly quota in about 3-4 days. Meaning I'm eating $20k worth of tokens for a $200 sub.
Current session I have running in the background, spinning up tests mostly
Total cost: $660.33
Total duration (API): 22h 36m 33s
Total duration (wall): 8h 33m 58s
Total code changes: 7662 lines added, 177 lines removed
Usage by model:
claude-haiku-4-5: 6.0k input, 38 output, 0 cache read, 0 cache write ($0.0062)
claude-opus-5: 322.9k input, 5.1m output, 702.9m cache read, 19.3m cache write ($606.64)
claude-opus-4-8: 742 input, 883.5k output, 38.6m cache read, 2.0m cache write ($53.68)