Check the Settings Developers Should Review Before Moving to Claude Sonnet 5.5
This page contains Coupang Partners affiliate links; the operator may earn a commission.
Key Takeaway
The first thing to check when moving to Claude Sonnet 5.5 (claude-sonnet-5-5) is the thinking setting. If thinking was turned off, switch to the new between_tools setting.
What are the price and model ID?
| Item | Value |
|---|---|
| Model ID | claude-sonnet-5-5 |
| Input | $2 per 1M tokens |
| Output | $10 per 1M tokens |
| Cache read | $0.20 per 1M tokens |
| Cache write | $2.50 per 1M tokens |
This is based on Anthropic's official announcement, checked on 2026-10-08. The unit prices are the same as Sonnet 5, and the cost drops because the same work takes fewer tokens, it says.
Are there settings in my code that I need to change?
Yes. The setting that keeps thinking turned off at the start is now between_tools. If you change only the model ID, requests may not behave as you expect, so check the migration guide first.
The default effort also differs by environment: Medium in Claude Code and the Claude app, and High on Claude Platform. Lowering it gives faster responses and uses fewer tokens, while raising it reasons longer and reviews results more carefully.
What security and account changes are there?
- High-risk cybersecurity work is handled by falling back to Sonnet 5. It says ordinary development work such as bug fixes is not affected, and a Cyber Verification Program is available for defensive users.
- A classifier that blocks extraction of reasoning content is applied for the first time. Preserved thinking is tied to the account that created it, so switching accounts mid-session may be restricted. The same holds in Claude Code, so set an operating rule to finish a given task under one account.
How should you read the benchmark numbers?
The announcement says it scored 70.6% on Terminal-Bench 4.0, well above Sonnet 5's 10.3%, and that on GDPval-AA it is 2 points below Opus 5.5. The provider sets the evaluation environment and conditions, so this does not mean the same numbers will reproduce on your codebase.
In what order should you migrate?
It is safer to leave the model string swap for last. First apply the thinking setting change, confirm that the default effort differs by environment, and then compare results against your existing model with a representative task set. Even with the same unit price, token usage can differ, so recalculate cost from actual billing.
Setting names and prices can change. Before deploying, reopen the model overview and migration guide in Anthropic's official documentation.
Frequently Asked Questions
Did the price go up?
Anthropic says input, output, and cache-read prices are the same as Sonnet 5. The explanation is that fewer tokens are used for the same work, so cost per task may come down, and the actual savings vary by workload.
Can I trust the performance numbers as they are?
The benchmarks in the announcement are values published by the provider. Before changing a production default, measure quality and cost again with your own task set.
Which clouds can I use it on?
The announcement says it is available on platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. Check each cloud console for availability by region.
Sources & Verification Date
- Introducing Claude Sonnet 5.5 (Anthropic, 2026-09-28) · Verified
- Newsroom (Anthropic) · Verified
All materials have been verified from official announcements and may have changed since verification.