跳到正文
DEV Community· Muhammad Moeed·· 7 天前精選AI 評分76

Claude Fable 5.2 與 Opus 5.5 比較:應該如何選擇?

Claude Fable 5.2 vs Opus 5.5: which one should you actually use?

AI 導讀

Claude Fable 5.2 與 Opus 5.5 的選用建議是,一般工作使用 Opus 5.5,較艱鉅的長程、多步驟任務才改用 Fable 5.2。

推薦理由

文章並列兩款模型的價格、基準表現與 Claude Code 切換方法,提供按任務難度分配模型的具體選用框架。

正文 · 原文

Muhammad Moeed

Anthropic released Claude Fable 5.2 on November 9, 2026. The first question everyone asks: should I switch from Claude Opus 5.5?

This is the short version. The full guide, with Claude Code setup and a longer FAQ, is on my blog: Claude Fable 5.2 vs Opus 5.5: Which One Should You Use?

Quick answer

  • Opus 5.5 for most work. Coding, bug fixes, writing, data analysis. $4 / $20 per million tokens.
  • Fable 5.2 for the hardest work. Large multi-file changes, long tasks the agent runs on its own, and bugs Opus keeps getting wrong. $8 / $40, exactly 2x Opus 5.5.
  • Anthropic's own launch post agrees: start with Opus 5.5 as your "daily driver" and keep Fable for "the most ambitious, open-ended technical work."

Side by side

Fable 5.2 Opus 5.5
Model ID claude-fable-5-2 claude-opus-5-5
Price (input / output) $8 / $40 $4 / $20
Cache reads $0.20 $0.20
Fast mode $16 / $80 $8 / $40
Context / max output 1M / 128K 1M / 128K
Default effort in Claude Code medium medium

Same context window, same output limit, same cache read price. Fable is not "bigger". It is better at long, multi-step work.

Where Fable 5.2 wins

From Anthropic's own benchmark table (not independent tests yet):

  • FrontierSWE v2 (agentic coding): 69.9% vs 62.3%
  • Remote Labor Index (long knowledge work): 41.3% vs 21.3%
  • Terminal-Bench-Science 0.1: 74.6% vs 58.7%
  • OSWorld 2.1 (computer use): 89.3% vs 87.2%, a small gap
  • AA-Briefcase Elo: Opus 5.5 is actually ahead, 1800 vs 1785

The pattern is clear: the bigger and longer the task, the bigger Fable's lead.

What 2x means in dollars

One task with 500K input tokens (400K of it cached) and 50K output tokens:

  • Opus 5.5: $0.40 + $0.08 + $1.00 = $1.48
  • Fable 5.2: $0.80 + $0.08 + $2.00 = $2.88

Run 20 of those every working day for a month (440 tasks) and it is $651 vs $1,267. That gap is fine if Fable gets hard tasks right the first time. It is wasted money on routine work.

One more detail: Opus 5.5 in fast mode costs the same as Fable 5.2 at normal speed. If you want speed, Opus fast mode is the cheaper path.

A simple rule

  1. Start on Opus 5.5 at the default effort.
  2. If the result is weak, raise effort to high, then xhigh.
  3. Still not good enough? Switch to Fable 5.2 for that one task.
  4. Switch back when the hard part is done.

Switching in Claude Code

Fable 5.2 needs Claude Code 2.1.321 or later:

claude update

Then:

/model fable

Go back with /model opus. One tip: in the /model picker, Enter saves your choice as the default for every new session, while s switches for this session only. Use s for Fable so you do not stay on the expensive model by accident.

Two API gotchas

  • Thinking is always on. Sending thinking: {type: "disabled"} or budget_tokens returns a 400. Use effort instead. Opus 5.5 works the same way.
  • Forced tool choice is gone. tool_choice set to any or a named tool returns a 400. Use auto and say in your prompt which tool to use.

來源:DEV Community · dev.to