Theo says a $200 plan buys $15,000 of Opus, $3,000 of Fable
A price cut on Fable did not turn into more work per subscription dollar, according to Theo's own request logs.
Theo put numbers on what a $200 Claude Ultra subscription is actually worth per model. By his math, the plan gets you about $15,000 of Opus 5.5 in a month but only around $3,000 of Fable. He says the figures come from analyzing over 1.1 million real requests to the Anthropic APIs on his own Ultra sub, continuing the request logging he has been running for a while.
If you exclusively use Fable, 50% of your limit will be "lost
The gap is not just the cap. Fable can only be used for 50% of a weekly limit, so a Fable-only workflow strands the other half. But Theo says that even setting the cap aside, Fable burns faster per dollar than Opus does.
Work done per week, not dollars per token
The measure he prefers is work done per week. If he used Opus exclusively, he reckons he could do roughly 1,200 turns and 230 threads in a week. If he exhausts the Fable limit and then switches to Opus, he gets about 900 turns, but slightly more threads. His explanation for the thread count is that Fable is more likely to finish work in a single prompt. He flags that a lot of his recent usage was benchmarking the new model, so the numbers may move.
The pricing did come down. Theo notes Fable 5.2 went from $10 in and $50 out to $8 in and $40 out, and says it also appears to get more output tokens per dollar through other efficiency wins, such as using fewer input tokens to reach a result.
That cut does not land the way the price sheet suggests. Addy Osmani, announcing Fable 5.2, said it generates output up to 30% faster than 5.1 and warned that it uses a lot of tokens. In Anthropic's own testing, he said, a given task tends to cost roughly the same as it did on 5.1 despite token pricing being 20% lower. He also recommended starting most work on Opus 5.5, calling it the strongest daily driver for coding and knowledge tasks, with Fable aimed at the most ambitious, open-ended technical work.
it costs roughly the same as Fable 5.1 even though the token pricing is 20% less
For working engineers, the useful takeaway is that per-token prices tell you very little about what a subscription buys. A model that spends more tokens to finish a task can be cheaper on the price sheet and more expensive in practice. Theo says these are early numbers and he will keep posting updates.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.