Opus 5.5 or Fable 5.1: what actually differs
2 min readAIClaudepricing
Two Claude models released three weeks apart. Compared not by the marketing but by the price sheet and by what Anthropic says about itself.
Fable 5.1 shipped on September 1, 2026. Opus 5.5 followed on September 22, the same month. Three weeks apart in one model family, and «which is better» turns out to split into two different questions: which is smarter and which is cheaper. The answers do not line up.
Price: a 2.5x gap, not a rounding error
Opus 5.5 costs $4 per million input tokens and $20 per million output tokens through the API. Fable 5.1 costs $10 and $50. These are not close numbers: Fable costs exactly 2.5 times more for the same volume of text, in both directions.
Caching widens the gap further. A cache hit on Opus 5.5 costs $0.20 per million tokens, on Fable 5.1 it is $0.25 - but the base a cache write is charged against starts higher on Fable regardless. For an agent holding a large context and revisiting it repeatedly, that is a difference in the monthly bill, not the per-token rate.
What Anthropic itself says
The company states it plainly: Opus 5.5 performs «at the level of Fable 5.1 on most work» while costing meaningfully less. Anthropic says a typical workload on Opus 5.5 runs roughly 40% cheaper than on the prior Opus 5, with output generated more than 30% faster.
That is the company's claim about its own product, and it is worth reading exactly that way - not as an independent measurement. The studio has no benchmark of its own to confirm or deny it, so saying that plainly is more honest than passing someone else's marketing number off as a verified fact.
Where Fable 5.1 stays a separate class
Fable and its Mythos relative are the line Anthropic positions as its most careful for complex work where the cost of a wrong answer is high: not speed or cost per token, but judgment quality in ambiguous situations. That is a separate axis from the «cheaper and faster» one Opus 5.5 is built around.
Agentic coding: a number worth checking yourself
On its own FrontierCode benchmark, Anthropic claims Opus 5.5 beats OpenAI's GPT-6 Astra at roughly 20% of the cost per task. That is the company's own number on its own test - compelling, and still not a reason to skip checking it against your own codebase before switching a working agent over.
What to pick
For high-volume agentic coding: Opus 5.5, at Opus-tier pricing with performance Anthropic says matches Fable. For work where one answer needs to be as careful as possible rather than as cheap as possible: Fable 5.1 stays a distinct option, and Anthropic is not charging two and a half times more for nothing.