I haven't typed /model fable in weeks. Opus 5.5 came out on September 22 and it's been so good that I never had a reason to switch. Before that I was manually adjusting my weekly Fable usage because I was using Fable so much.
That's a strange thing to say about Fable 5.1, which is supposed to be Anthropic's top model. When Fable 5 launched in June, people fought over access to it, and we wrote about how it came back to Max plans capped at 50 percent. Four months later, the model one tier down is beating it on Anthropic's own scoreboard.
Opus 5.5 wins on every row Anthropic published
Anthropic's Opus 5.5 announcement put the two side by side across nine benchmarks, and Opus 5.5 scores higher on all nine. Some of the gaps are small. Others aren't:
- Terminal-Bench 4.0, agentic coding: 66.4 percent for Opus 5.5, 55.8 percent for Fable 5.1
- CursorBench 4.0: 57.8 percent versus 51.8 percent
- AutomationBench, business workflows: 40.0 percent versus 31.4 percent
- Humanity's Last Exam with tools: 67.7 percent versus 65.6 percent
- OSWorld 2.1, computer use: 81.8 percent versus 80.7 percent
The announcement says Opus 5.5 "performs at the level of Claude Fable 5.1 on most work." On one long rewrite job Anthropic ran, "Opus 5.5 finished in 9.5 hours compared to 12 for Fable 5.1, and cost 51% less." And on CursorBench, Opus 5.5 at its default medium effort scored 52.5 percent, slightly above Fable 5.1 running at max.
And Opus is a lot cheaper to use. Per Anthropic's model overview, Fable 5.1 costs $10 per million input tokens and $50 per million output tokens. Opus 5.5 is $4 and $20. That's 2.5 times the price per token for a model that loses the comparison. Opus 5.5 is also faster, and it's the default in Claude Code on every paid plan, while the docs say plainly that "Fable models are not the account-type default on any plan or provider." You have to go out of your way to use Fable, and right now there isn't much reason to.
The same announcement also admits "the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest." Its model guide still recommends Fable 5.1 "for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short." And in Every's review of Opus 5.5, the team noted that "For the hardest coding problems, Mike and Kieran still reach for Fable."
So Fable still has a job. It's just a narrow one: the hardest few percent of problems, where Opus 5.5 at high effort gets stuck. For a flagship, that's not much of a pitch.
What would bring me back
A top model should be worth reaching for. Here's what would get me to switch back, roughly in order.
Make it clearly smarter again. A Fable that only ties Opus is a Fable nobody picks. The next version needs a gap people can feel on real work, the way Fable 5 felt in June.
Close the price gap. 2.5 times the cost is easy to justify when the model is clearly better and hard to justify when it isn't. Even halving the difference would change the math for a lot of API users.
Give subscribers real room to use it. If Fable only fits in a slice of your plan, you save it for emergencies and stop building habits around it. That's how a model gets forgotten.
Keep it from tripping over itself. Fable 5's safety filters were the first thing people complained about, and while Fable 5.1 improved that, a model that hands some requests off to a different one is harder to trust with the hard work it's supposed to be for.
None of this is a complaint about Opus 5.5, which is the best everyday Claude I've used, and which brought back the feel of Opus 4.6. It's that a lineup only works when each tier has a reason to exist. Right now the top tier doesn't, at least not for me. As of mid-October, Fable 5.1 is still the newest model at the top of Anthropic's own model list.
Sources
Anthropic: Introducing Claude Opus 5.5 - The benchmark table comparing Opus 5.5 and Fable 5.1 across nine evaluations, the statement that Opus 5.5 performs at the level of Fable 5.1 on most work, the 9.5-hour versus 12-hour rewrite at 51 percent lower cost, the CursorBench default-effort comparison, and the note that the real-world gap is narrower than the scores suggest.
Anthropic: Models overview - Fable 5.1 priced at $10 and $50 per million input and output tokens and Opus 5.5 at $4 and $20, and the guidance to use Fable 5.1 for demanding long-horizon work or when Opus 5.5 at higher effort falls short.
Claude Code docs: Model configuration - Opus 5.5 as the default on paid plans and the statement that Fable models are not the default on any plan or provider.
Every: Vibe Check on Opus 5.5 - The note that the team still reaches for Fable on the hardest coding problems.
