I've had an extra-usage quota for months now. They even gave people free credits for that some time ago.
It's useful once in a while when you're like 90% done, run out of usage and the next refresh is hours away. Then you can push that feature to completion and swap to another model that's not on cooldown.
It's just matter of time that price become so high that Enterprise cannot afford these tools unless they self-host open models. Only problem is open models are coming from China and not many countries trust that to use inside companies. How will it play?
Looks like the Chinese are, again, winning the long game.
There is simply nothing that could compete with their open models. At the same time more and more corps got "AI addicted", so they will either have to pay ridiculous amounts of money, or use the Chinese stuff.
You are right about addiction; the instant gratification is on peak with tools like Claude. I use $200 plan and my token usage for last 30 days is 30M , If I do basic math, it will be equivalent to $1500. Assuming companies get 50% discount, still that cost is too much multiplier.
My usage might be outliner here, but 200->400 is an average case.
At least for coding, there's little correlation between token spend and the quality (and impact) of the resulting AI suggestion.
This is fine when inference prices are capped (eg via a monthly subscription plan or self-hosting), but rapidly discombobulates the relationship between provider and user otherwise.
It still seems like OpenAI has no moat and neither does anyone else, as the only reasonable way to use the coding slot machines is going to be via open source models on inference-optimized hardware.
Still better than the secret lobotomization they were doing on subscription plan models though.
>there's little correlation between token spend and the quality
My sentiment exactly! I have a very similar scaffold to each of my prompts, and feel I provide similar context files, however sometimes I get a truly inspired, complex, and functionally complete response...and sometimes I'd have been better off running lorem ipsum through a python interpreter.
I can't find any rhyme or reason to success. I'm not sure if prompting is significantly more nuanced than I realise, or it's the statistical magic that's having a laugh at me.
>open source models on inference-optimized hardware.
Is this actually a thing? Or are you talking about some hypothetical "opus 4.7 ASIC"?
Anthropic is doing changes on their help support pages on what looks like it will be the next pricing change regarding how users will use Opus models on Pro Plan.
OpenAI and Anthropic are both planning IPOs this year. They are clearly trying to polish their finances before filing their S-1s. Because their advisors will have told them that it's going to be a very difficult sell at these valuations if they cannot at least present the idea of a path towards profitability.
i've been using the API for a portion of my work to prepare for this and test to see how much it will cost in the long term, turns out that was the short term.
it is rough, but it has taught me to treat every prompt and process with care since i watch the pennies and dollars burn instead of tokens, which is a good habit to get into anyway
Copilot has said they'll be giving out previous months "if you did token pricing, it would cost $x" so a lot of us will have real numbers to actually anticipate our spend.
Personally I'm anticipating agentic coding will be out of my price bracket (a single agent run costing US$20+ is far beyond what I can justify, especially with how often it fails). I'm planning on going back to optimised prompts on one-pass edits.
I use pi (https://pi.dev/) as my coding agent which allowed you to use your subscription until Claude recently introduced extra usage and I was routinely using the equivalent of my Claude Max subscription in an hour multiple times a day.
The messaging around what is and isn't allowed with the various Claude plans has been so very muddled as of late. Add to that declining model performance, changes to default reasoning efforts, expanded token usage, caching bugs, corporate denials and gaslighting -- I don't think it's overstating matters to say they've suffered some major self-inflected reputational damage.
As it stands now, there is so much FUD surrounding their offerings, I'm not sure what they could do in the short term to turn things around.
They need to start shifting from "move fast and break things" to "move faster by slowing down". Their public communication, feature set, and organization as a whole needs to start matching the scale and level they're competing at. They won many hearts and minds by being better and are losing them by being chaotic. Different outcomes from the same internal behavior because they needed to change gears and haven't.
Investor funds have been subsidizing the inference costs so far.
Investors might move from funding the model providers to funding the enterprises that use those models. That is, they might move from funding the cost of the experiment to funding the value of the result. No funding if there are no demonstrable AI gains.
This is a reasonable shift if this happens. If enough gains have been demonstrated, then investors might go back to funding the model providers. Investors always move towards the highest leverage point.
> this is an outdated support article from before we had Opus in Pro plans, we rolled it out for Opus 4.5 in Jan and never updated, you can see in the wayback machine here: https://web.archive.org/web/20251204151142/https://support.c...
> apologies for the confusion
response from @thariq
I don't see any evidence of that, the current, live page clearly states:
> Model access: When using a Pro plan with Claude Code, you will only be able to use Opus models after enabling and purchasing extra usage.
And your extremely slow loading archive link is just the same page with the same text.
> I don't see any evidence of that, the current, live page clearly states
The live page - last edited 15 minutes after your comment - no longer has that passage.
Thank-you for posting this update. EDIT: However purchasing extra-usage for Pro/Max accounts is a new feature.
I've had an extra-usage quota for months now. They even gave people free credits for that some time ago.
It's useful once in a while when you're like 90% done, run out of usage and the next refresh is hours away. Then you can push that feature to completion and swap to another model that's not on cooldown.
It's been a feature on the Max accounts for a while now.
Their vibe-coded "documentation" sucks so much…
I thought documentation is a solved problem and you can just have an agent keep all your docs up-to-date. /s
You have one agent write the code, a separate one to do the docs, a third to harmonise them.
After a few weeks they'll settle on the documentation for raised garden bed and the implementation of a home defence sprinkler system.
All while leaving you with a $10,000 bill at the end of the month.
Ain't life grand.
It's just matter of time that price become so high that Enterprise cannot afford these tools unless they self-host open models. Only problem is open models are coming from China and not many countries trust that to use inside companies. How will it play?
Looks like the Chinese are, again, winning the long game.
There is simply nothing that could compete with their open models. At the same time more and more corps got "AI addicted", so they will either have to pay ridiculous amounts of money, or use the Chinese stuff.
You are right about addiction; the instant gratification is on peak with tools like Claude. I use $200 plan and my token usage for last 30 days is 30M , If I do basic math, it will be equivalent to $1500. Assuming companies get 50% discount, still that cost is too much multiplier.
My usage might be outliner here, but 200->400 is an average case.
Many people trust China more than US
US investors can't afford price being too high tho. This thing would wipe trillions if they make prices a lot higher. Interesting times ahead
Yep, it is their responsibility to make all of this worth caring about.
Hence the only other path: slow, constant enshittification of the models' performance
[dead]
Opus was consuming so much usage that it was basically unusable on anything but the Max plan.
It was even burning out much too fast on Max, but I found moving from xhigh -> high thinking got me equivalent quality with fewer tokiens.
This is going to be interesting.
At least for coding, there's little correlation between token spend and the quality (and impact) of the resulting AI suggestion.
This is fine when inference prices are capped (eg via a monthly subscription plan or self-hosting), but rapidly discombobulates the relationship between provider and user otherwise.
It still seems like OpenAI has no moat and neither does anyone else, as the only reasonable way to use the coding slot machines is going to be via open source models on inference-optimized hardware.
Still better than the secret lobotomization they were doing on subscription plan models though.
>there's little correlation between token spend and the quality
My sentiment exactly! I have a very similar scaffold to each of my prompts, and feel I provide similar context files, however sometimes I get a truly inspired, complex, and functionally complete response...and sometimes I'd have been better off running lorem ipsum through a python interpreter.
I can't find any rhyme or reason to success. I'm not sure if prompting is significantly more nuanced than I realise, or it's the statistical magic that's having a laugh at me.
>open source models on inference-optimized hardware.
Is this actually a thing? Or are you talking about some hypothetical "opus 4.7 ASIC"?
Basically the title.
Anthropic is doing changes on their help support pages on what looks like it will be the next pricing change regarding how users will use Opus models on Pro Plan.
Turns out when the bills start rolling in you need the revenue to pay them. Here comes the bumpy ride
OpenAI and Anthropic are both planning IPOs this year. They are clearly trying to polish their finances before filing their S-1s. Because their advisors will have told them that it's going to be a very difficult sell at these valuations if they cannot at least present the idea of a path towards profitability.
A few years from now we’ll probably look back and laugh at how AI used to be essentially free.
i've been using the API for a portion of my work to prepare for this and test to see how much it will cost in the long term, turns out that was the short term.
it is rough, but it has taught me to treat every prompt and process with care since i watch the pennies and dollars burn instead of tokens, which is a good habit to get into anyway
Copilot has said they'll be giving out previous months "if you did token pricing, it would cost $x" so a lot of us will have real numbers to actually anticipate our spend.
Personally I'm anticipating agentic coding will be out of my price bracket (a single agent run costing US$20+ is far beyond what I can justify, especially with how often it fails). I'm planning on going back to optimised prompts on one-pass edits.
I use pi (https://pi.dev/) as my coding agent which allowed you to use your subscription until Claude recently introduced extra usage and I was routinely using the equivalent of my Claude Max subscription in an hour multiple times a day.
The writing is on the wall.
The messaging around what is and isn't allowed with the various Claude plans has been so very muddled as of late. Add to that declining model performance, changes to default reasoning efforts, expanded token usage, caching bugs, corporate denials and gaslighting -- I don't think it's overstating matters to say they've suffered some major self-inflected reputational damage.
As it stands now, there is so much FUD surrounding their offerings, I'm not sure what they could do in the short term to turn things around.
It's just an organizational maturity thing.
They need to start shifting from "move fast and break things" to "move faster by slowing down". Their public communication, feature set, and organization as a whole needs to start matching the scale and level they're competing at. They won many hearts and minds by being better and are losing them by being chaotic. Different outcomes from the same internal behavior because they needed to change gears and haven't.
Investor funds have been subsidizing the inference costs so far.
Investors might move from funding the model providers to funding the enterprises that use those models. That is, they might move from funding the cost of the experiment to funding the value of the result. No funding if there are no demonstrable AI gains.
This is a reasonable shift if this happens. If enough gains have been demonstrated, then investors might go back to funding the model providers. Investors always move towards the highest leverage point.
As long as AI delivers, this would be the rhythm.
Investors can try to subsidize while the cost of delivery, evolution, and efficiency improves, while at the same time completing market capture.