Opus 5.5 vs Fable

Overview

Both are Anthropic high-end models with 1M context. Opus 5.5 is the default “hard task” SKU at $4 / $20. Fable 5.1 is slower, adaptive thinking always on, and $10 / $50—about 2.5× Opus on both input and output. Start on Opus; use Fable when the extra spend is the point.

Capabilities

• Opus 5.5 — hard cases at mid-premiumArchitecture, messy bugs, and multi-constraint specs without Fable’s 2.5× token bill.
• Fable 5.1 — max Anthropic qualityWhen you would rather pay $10 / $50 than miss a subtle requirement. Treat it as a review / final-pass model.
• Same family, different spendPrompts transfer. Don’t rewrite the brief—change the SKU after Opus has framed the problem.

Context Window

Both list 1M input and 128K max output. Fable in a huge prompt is expensive: 1M input at list is $10 vs $4 on Opus.

• Escalate the recap, not the whole repo history.
• Fable cache hits are cheaper as a share of input (2.5%) but the base input is still $10.
• Check Anthropic’s current cards before you assume windows.

Pricing

Anthropic list prices. Hosts may differ. Fast mode on Opus (where offered) is a premium on top of $4 / $20.

Opus 5.5

$4 in / $20 out · 5m cache write $5 · 1h $8

Fable 5.1

$10 in / $50 out · 5m cache write $12.50 · 1h $20

Rule of thumb

Fable ≈ 2.5× Opus on in and out

Anthropic pricing

How to Use

1. Draft on Opus 5.5Plan and first patch at $4 / $20.
2. Promote to Fable if stuckPass the recap, failing test, and constraints—not the whole thread.
3. Don’t default FableA day of agent loops at $10 / $50 adds up faster than Opus.

Alternatives

• Fable 5.1 vs Opus 5.5Same pair, Fable-first framing.
• Sonnet 5.5Daily driver at $2 / $10.
• Opus 5.5 vs AstraAnthropic Opus vs OpenAI GPT-6 Astra.