Claude Sonnet 5.5
Claude Sonnet 5.5 is Anthropic's September 28, 2026 Sonnet release. Use claude-sonnet-5-5 on the Claude API; Anthropic documents 1M context, 128K maximum output, text and image input, text output, adaptive thinking, and $2 input/$10 output per million tokens.
Why it matters
Sonnet 5.5 gives teams a current lower-priced option beside Opus 5.5 for well-scoped coding and agent tasks. Its $2/$10 list token rates match the current Sonnet 5 rates; Anthropic's faster and lower task-cost claims are its own measurements, so test the migration with your prompts and acceptance checks.
Current model and API contract
The Claude API model ID is claude-sonnet-5-5. Anthropic lists a 1M-token context window, 128K maximum output, text and image input, text output, adaptive thinking, high default effort, and a June 2026 reliable knowledge cutoff. The model page lists Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS; account-level access still depends on the platform.
Price and task-cost boundary
Claude API base rates are $2 input and $10 output per million tokens, matching the current Sonnet 5 rates. Cache reads are $0.20, five-minute writes $2.50, and one-hour writes $4 per million. Anthropic reports lower task cost and faster generation in its own testing; these are not token-rate discounts or independent measurements.
Migration checks before switching
A Sonnet 5 application may need changes: use between_tools to disable up-front thinking, remove forced tool use, preserve thinking blocks with the conversation, replace the older computer-use tool on the Claude API or Google Cloud, and check advisor-model support. Streaming clients should also handle text between tool calls appearing in thinking blocks. Anthropic's migration guide documents these changes; run your own regression tests before changing a production route.
When to choose Sonnet or Opus
Anthropic positions Sonnet 5.5 for bounded everyday coding and polished document work, while Opus 5.5 remains its recommended starting point for most workloads and stronger choice for complex open-ended work. Try both at a defined effort on representative tasks, then compare accepted results and total cost rather than assuming one launch benchmark decides the routing.
Useful first jobs
- Everyday code fixes and implementation with verification.
- Cost-aware agent workflows with tools and image input.
- Document, slide, and spreadsheet drafting with review.
- Migration trials from Claude Sonnet 5.
- Routing bounded tasks against Claude Opus 5.5.
Frequently asked questions
What is the Claude Sonnet 5.5 API model ID?
Use claude-sonnet-5-5. The current Anthropic model page lists 1M context, 128K output, text and image input, and adaptive thinking.
Is Sonnet 5.5 30% cheaper per token than Sonnet 5?
No. Anthropic lists $2 input and $10 output per million tokens for Sonnet 5.5 and reports up to 30% lower cost per task in its own tests. Your task cost depends on token use, cache hits, retries, and result quality.
Can I switch from Sonnet 5 without code changes?
Check the migration guide first. Thinking controls, forced tool use, preserved thinking blocks, computer-use tools, advisor support, and streaming text behavior can require changes.
Sources and evidence
Anthropic's release, live model page, and migration guide establish identity, pricing, limits, platform list, and breaking API behavior. Anthropic's benchmark and task-cost comparisons are provider measurements.
- Introducing Claude Sonnet 5.5 ↗
Anthropic · Primary source
- Claude Sonnet 5.5 model page ↗
Anthropic · Primary source
- Migrating to Claude Sonnet 5.5 ↗
Anthropic · Primary source
- Claude API pricing ↗
Anthropic · Primary source