Anthropic released Fable 5.2 on November 9, almost seven weeks after Opus 5.5. Fable is clearly the flagship again, even if its version number looks behind the rest of the lineup. Both Anthropic’s evaluations and Artificial Analysis’s results show it is the more capable model. It also comes at a premium, and that is where I am less convinced.

Back when Fable 5 and Opus 5 were released, the sequence made sense. Fable 5 arrived in June as Anthropic’s most capable generally available model. Opus 5 followed in July, bringing comparable intelligence at half the token price. Artificial Analysis found them effectively tied overall, with Opus ahead on its professional-work benchmarks. The flagship came first, and then a more affordable model caught up. That was an easy progression to understand, even if it made Fable harder to justify.

This time, Fable not only arrived after Opus but still trails it in version number. On Artificial Analysis’s Intelligence Index, Fable 5.2 scores 65 against Opus 5.5’s 59, both at maximum effort. Fable’s API prices are $8 per million input tokens and $40 per million output tokens, twice Opus’s $4 and $20. On that benchmark, the actual cost per task is $26.54 against $10.04. Six points buys a measurable lead, but the bill grows considerably faster than the score.

A case can be made for specialized work. Some of Anthropic’s reported gains are much larger than the overall Index gap, and by its own figures, a lower-effort Fable can match Opus’s best result more cheaply on certain tests. Yet Anthropic still recommends starting with Opus 5.5 for most coding and knowledge work. I can accept leaving general planning to Opus and bringing in Fable for specialized workloads. My trouble with that rule is that most models have specialized cases where they are great. It is harder to find one that doesn’t.

For Pro subscribers, this is at least a good week to try it. Fable is included through November 16, for up to half the weekly allowance. Two banked resets have also been issued: one general reset and one for Fable usage. That gives subscribers room to see what the extra capability does on their own work.

More choices are a plus for Claude users. Fable 5.2 gives people a higher ceiling. But finding it a specialized use is much easier than convincing me to absorb the premium. Six extra points on a synthetic benchmark have not convinced me to pay twice the token price.