I cancelled my subscription two days ago. I was a customer since October last year. I could get a decent bit of work done on just the 20$ subscription, but since this monday, I can barely get two prompts in before hitting my limit.
Same codebase, same sort of prompt, same scale. I was already on the fence. Models like Qwen, Kimi, or GLM5 already go a very long way while being vastly cheaper, and the new openAI models feels equivalent but with higher limits.
This is getting to the point where the right harness makes a bigger difference than the right model. I've been experimenting with some planner-executor-reviewer setup in opencode, and I'm starting to feel like multiple smaller models working together are netting me better results.
With the introduction of Opus 4.6 my bills went through the roof. I never burned budgets so fast with so few prompts since then.
I more and more use Codex, because token usage is a blackbox and I think that we will see the next couple of month the usual three tier model evolving: free, normal, luxury.
2027 will be the year of token regulation by administrations worldwide. Until then take care for being ripped of at the luxury level.
I love the interface of Claude Code but with these limits I’d be willing to use Codex. Anyone know if it’s possible to use Claude Code with other provider subscriptions (not API usage costs)?
Whilst you need to generate an API Key, you are still using the 'GLM Coding Plan' as per 'Methods for Using the GLM Coding Plan in Claude Code' documentation.
First, this has been going on for a week now and Anthropic support was gaslighting people, saying nothing has changed. Well, obviously it has, they said it now. Additionally, Anthropic saying weekly limits are unaffected is also a lie. I exhausted 37% of my weekly on the first day when this started on Monday. This has never happened before. So this is two lies in a row.
But now the bigger issue - what does Anthropic expect me to do when I'm out of quota on day 1 of the week? Emphasis on me having a lot of free time without access to claude.
There can only be 3 options:
1. I'll buy more $200 subscriptions from a provider that I've caught lying and gaslighting me.
2. I'll only work 1 day in a week and accomplish nothing.
3. I'll explore other models / providers to see if I can fill the gap.
The right choice is very clear. But can't Anthropic see that their only moat is that ecosystem was built around claude code because they were first to do this subscription stuff and their harness didn't get in the way too much?
From carrot to stick huh. Not enough people took up the off peak bonus I'm guessing, so make on peak a penalty. Kinda sucks, but be nice to have a bonus during the really dead hours to compensate.
Anthropic is probably the only AI company that is trying to stop absolutely hemorrhaging money and reach profitability and financial sustainability.
They aren't far off - they burn a tiny fraction of the cash of OAI and achieve similar ARR despite this - but as they tighten the belt it's inevitable that companies like OAI come in and offer more subsidized (unsustainable) inference to get people to switch. They will inevitably do the same "rug pull".
The winner will be the last man standing. The company which is able to continue subsidizing tokens after all others have been forced to throttle and raise prices. Soak up the market share, wait for your rivals’ investors to get cold feet, then raise prices the morning after the last one folds. (It’s Google it’s always been Google)
To this day, nobody has proven any subsidization is actually taking place. The only "proof" is comparing subscription with API prices, but that's comparing two very different products. Unrestricted enterprise APIs always cost more than consumer products. Sometimes a whole lot more, so API prices don't really tell you much in terms of inference costs.
And there's a reason why Anthropic cracked down on third party harnesses - prompt caching must be very effective for cost reduction, and if you look at cache hits of CC, you'll see it's very aggressive. They even said it themselves on twitter that if your limits get exhausted sooner, it's likely something with your workflow that doesn't utilize caching as well.
If any subsidization is taking place, I'd wager it's simply light users subsidizing heavy ones. But that's nothing new.
This will just get worse, particularly when you consider just how shitty the US government is becoming at properly managing the basic necessities for stability.
IF you arn't planning a local LLm strategy, you're surely tying your lifeline to anchors.
Computers are getting more and more expensive. If this continues, one day we'll be priced out of computers. There won't be a way to run local models. At least not one peasants like us can afford.
Funny how before the announcement, people who were experiencing this were being gaslighted on different platforms, to think they have a "skill" issue using Claude Code
Additionally, this was practically predicted and expected by so many people, the second the off-hours increase was announced.
Its a sign that they cannot simply raise the price of existing paying customer's to avoid putting on limits - and/or - convert more free users into paying customers.
The longer this goes on the more it becomes clear Google is going to be the last one standing.
Why because 7% of people complain the loudest? I would imagine that most power users have big followings on X, Reddit, HN, etc.. but that hardly reflects the reality of what most people are experiencing.
I cancelled my subscription two days ago. I was a customer since October last year. I could get a decent bit of work done on just the 20$ subscription, but since this monday, I can barely get two prompts in before hitting my limit.
Same codebase, same sort of prompt, same scale. I was already on the fence. Models like Qwen, Kimi, or GLM5 already go a very long way while being vastly cheaper, and the new openAI models feels equivalent but with higher limits.
This is getting to the point where the right harness makes a bigger difference than the right model. I've been experimenting with some planner-executor-reviewer setup in opencode, and I'm starting to feel like multiple smaller models working together are netting me better results.
With the introduction of Opus 4.6 my bills went through the roof. I never burned budgets so fast with so few prompts since then.
I more and more use Codex, because token usage is a blackbox and I think that we will see the next couple of month the usual three tier model evolving: free, normal, luxury.
2027 will be the year of token regulation by administrations worldwide. Until then take care for being ripped of at the luxury level.
I love the interface of Claude Code but with these limits I’d be willing to use Codex. Anyone know if it’s possible to use Claude Code with other provider subscriptions (not API usage costs)?
You can!
According to the GLM documentation at least, you can edit you .env (~/.claude/settings.json) with an API key from the GLM Settings page. [0].
{ "env": { "ANTHROPIC_AUTH_TOKEN": "your_zai_api_key", "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic", "API_TIMEOUT_MS": "3000000" } }
Whilst you need to generate an API Key, you are still using the 'GLM Coding Plan' as per 'Methods for Using the GLM Coding Plan in Claude Code' documentation.
[0] - https://docs.z.ai/devpack/tool/claude
Yes, I've run Claude Code using ollama running on my GPU server with a 5090.
Not sure what the plan is here to be honest.
First, this has been going on for a week now and Anthropic support was gaslighting people, saying nothing has changed. Well, obviously it has, they said it now. Additionally, Anthropic saying weekly limits are unaffected is also a lie. I exhausted 37% of my weekly on the first day when this started on Monday. This has never happened before. So this is two lies in a row.
But now the bigger issue - what does Anthropic expect me to do when I'm out of quota on day 1 of the week? Emphasis on me having a lot of free time without access to claude.
There can only be 3 options:
1. I'll buy more $200 subscriptions from a provider that I've caught lying and gaslighting me. 2. I'll only work 1 day in a week and accomplish nothing. 3. I'll explore other models / providers to see if I can fill the gap.
The right choice is very clear. But can't Anthropic see that their only moat is that ecosystem was built around claude code because they were first to do this subscription stuff and their harness didn't get in the way too much?
From carrot to stick huh. Not enough people took up the off peak bonus I'm guessing, so make on peak a penalty. Kinda sucks, but be nice to have a bonus during the really dead hours to compensate.
Anthropic is probably the only AI company that is trying to stop absolutely hemorrhaging money and reach profitability and financial sustainability.
They aren't far off - they burn a tiny fraction of the cash of OAI and achieve similar ARR despite this - but as they tighten the belt it's inevitable that companies like OAI come in and offer more subsidized (unsustainable) inference to get people to switch. They will inevitably do the same "rug pull".
It'll be interesting to see how this plays out.
The winner will be the last man standing. The company which is able to continue subsidizing tokens after all others have been forced to throttle and raise prices. Soak up the market share, wait for your rivals’ investors to get cold feet, then raise prices the morning after the last one folds. (It’s Google it’s always been Google)
I think a couple Chinese companies will be up here. And they likely won't raise at all, because they don't need to.
This feels like the natural end of the subsidized tokens phase.
Once everyone tightens limits, the competition shifts from model quality to pricing strategy and usage control.
To this day, nobody has proven any subsidization is actually taking place. The only "proof" is comparing subscription with API prices, but that's comparing two very different products. Unrestricted enterprise APIs always cost more than consumer products. Sometimes a whole lot more, so API prices don't really tell you much in terms of inference costs.
And there's a reason why Anthropic cracked down on third party harnesses - prompt caching must be very effective for cost reduction, and if you look at cache hits of CC, you'll see it's very aggressive. They even said it themselves on twitter that if your limits get exhausted sooner, it's likely something with your workflow that doesn't utilize caching as well.
If any subsidization is taking place, I'd wager it's simply light users subsidizing heavy ones. But that's nothing new.
This will just get worse, particularly when you consider just how shitty the US government is becoming at properly managing the basic necessities for stability.
IF you arn't planning a local LLm strategy, you're surely tying your lifeline to anchors.
Computers are getting more and more expensive. If this continues, one day we'll be priced out of computers. There won't be a way to run local models. At least not one peasants like us can afford.
Funny how before the announcement, people who were experiencing this were being gaslighted on different platforms, to think they have a "skill" issue using Claude Code
Additionally, this was practically predicted and expected by so many people, the second the off-hours increase was announced.
Shoddy company
Its a sign that they cannot simply raise the price of existing paying customer's to avoid putting on limits - and/or - convert more free users into paying customers.
The longer this goes on the more it becomes clear Google is going to be the last one standing.
Why because 7% of people complain the loudest? I would imagine that most power users have big followings on X, Reddit, HN, etc.. but that hardly reflects the reality of what most people are experiencing.
The remaining 93% exist mostly because these 7% were loud.
Users who aren't using their quota will gradually disappear when that 7% starts being loud in the other direction.
that explains why i kept hitting limits this week