This should be discussed more often. LLM providers can simply degrade model quality or reduce quotas and there is absolutely nothing you can do about it, making subscription price meaningless.
There should be a fixed equivalence to APIs, like "monthly subscriptions at tier X get the equivalent of 20 monthly sessions of API Y, up to Z tokens".
Instrumented monitoring of Anthropic's rate limit headers reveals inconsistent quota consumption rates that cannot be explained by the "holiday bonus expiration" cited by Anthropic. The same user, same plan, same workload type shows burn rates varying from 5.6%/hour to 59.9%/hour within the same 48-hour period.
This 10x variance constitutes either a bug in quota accounting or an undisclosed server-side change to rate limiting behavior.
Interesting, not sure if that's enough of a sample size, but the argument of "this is unreasonable for a 200$/month consumer subscription" will make or break all kinds of services.
Nice work! I would take into account how slow the shi..nice CLI of CC is!
It's not only the slow responses or other means of making the work slower in the background (for anthropic means slowing down pure money saving -> big incentive), it's also not fixing their CC client, leaving it slow like a snail. Key presses are lost making corrections necessary. But you have to wait until CC finishes it's ongoing exploration/update/writing/ bash.. Tool calls are slow as hell. Everything is slow!
It's like to buy a Ferrari and the manufacturer pays it's fuel per minute of usage - then make the Ferrari starting up very slowly, make it react to input after 10-30 seconds, let it artificialy slow down everything.
Management Task: "promise, and then hide adverse actions without transparency or possibility to control for the driver!" -> task solved very effectively, bonus saved!
I really hope some will take it to court, in California, in Europe and every other place affected.
There should be a fixed equivalence to APIs, like "monthly subscriptions at tier X get the equivalent of 20 monthly sessions of API Y, up to Z tokens".
This 10x variance constitutes either a bug in quota accounting or an undisclosed server-side change to rate limiting behavior.
It's not only the slow responses or other means of making the work slower in the background (for anthropic means slowing down pure money saving -> big incentive), it's also not fixing their CC client, leaving it slow like a snail. Key presses are lost making corrections necessary. But you have to wait until CC finishes it's ongoing exploration/update/writing/ bash.. Tool calls are slow as hell. Everything is slow!
It's like to buy a Ferrari and the manufacturer pays it's fuel per minute of usage - then make the Ferrari starting up very slowly, make it react to input after 10-30 seconds, let it artificialy slow down everything.
Management Task: "promise, and then hide adverse actions without transparency or possibility to control for the driver!" -> task solved very effectively, bonus saved!
I really hope some will take it to court, in California, in Europe and every other place affected.
Keep up vibe coding! Hahhaha