What's this about?
Within three hours of launch, users reporting that Fable 5.1 exhausts the five-hour window in under thirty minutes — from both 5× and 20× subscribers. dotey's version: you hit the five-hour limit quickly even on Max@20, your Fable allowance goes to 100%, and you are left with "chicken-rib" Opus 5 — a Chinese idiom for something not worth eating and a shame to throw away.
Why would a better model burn quota faster?
Because quota is metered in tokens, and a model that reasons longer before answering spends more of them. Anthropic notes Fable 5.1 defaults to High effort in Claude Code and Medium elsewhere, and dotey reports Max effort costs 1.5× or more. So the same subscription buys fewer of the more expensive units. The price reduction is real and applies to API cache reads; it does not change how a subscription meters you.
What's the community saying?
- Anthropic reset both 5-hour and weekly limits on launch day, which blunted the complaint for a day and did not change the rate.
- The counter-intuitive complaint from HN: the reset itself hurt some users by pushing their nearly-expired 5-hour window back to a full five hours.
- dotey's practical mitigations are worth more than the complaint: disable the 1M context window, avoid Max effort unless needed, and be sceptical of the official advice to skip `/compact`.