Sonnet 5.5 vs Opus 5.5

Overview

Same Claude 5.5 generation, two prices. Sonnet 5.5 is $2 / $10 and fast—the default. Opus 5.5 is $4 / $20 (2× on both sides) with stronger hard-case reasoning. Start on Sonnet; escalate a recap to Opus when Sonnet loops.

Capabilities

• Sonnet 5.5 — volumeAgent loops, PR-sized diffs, docs. Output at $10 / MTok still dominates cost if you print whole files.
• Opus 5.5 — stuck or subtlePay 2× when the design is open, constraints conflict, or Sonnet keeps missing a requirement.
• Same cache-read dollarBoth cache reads list at $0.20 / MTok. Opus writes cost more to fill the cache.

Context Window

Both are 1M / 128K class on Anthropic’s cards. Doubling the prompt on Opus doubles a higher base rate.

• Cache the prefix on Sonnet for the whole session.
• Send Opus a short recap + evidence, not 400K of logs.

Pricing

Opus 5.5 is 2× Sonnet 5.5 on input and output ($4 / $20 vs $2 / $10)

Anthropic list. Sonnet 5.5 matches Sonnet 5’s $2 / $10 standard rate. Confirm on Anthropic’s pricing page.

Sonnet 5.5

$2 in / $10 out · cache read $0.20 · batch $1 / $5

Opus 5.5

$4 in / $20 out · cache read $0.20 · batch $2 / $10

50K in + 2K out

Sonnet ≈ $0.12 · Opus ≈ $0.24

Anthropic pricing

How to Use

1. Default Sonnet 5.5`claude-sonnet-5-5` for the first plan and patch.
2. Escalate onceIf tests still fail, one Opus 5.5 call with the recap.
3. Come back downAfter Opus unblocks, continue on Sonnet so the rest of the day stays at $2 / $10.

Alternatives

• Sonnet 5.5 costFull rate card: cache, batch, worked example.
• Opus 5.5 vs FableNext step up from Opus.
• GPT-6 Sol vs LunaOpenAI’s $2 vs $0.10 mid/low tier.