Claude Sonnet 5.5: API pricing, benchmarks, and coding guide
A practical look at Anthropic's faster Sonnet model: published benchmarks, API economics, coding and knowledge-work strengths, safety changes, and the migration checks that matter in production.
Release summary
What is Claude Sonnet 5.5?
Anthropic announced Claude Sonnet 5.5 on as the second model in the Claude 5.5 family. It is designed as a faster, lower-cost complement to Opus 5.5: strong on well-scoped everyday work, bug fixes, coding, documents, slides, spreadsheets, image understanding, and long-running tasks.
The practical upgrade is not only a higher score. Anthropic reports output generation that is more than 30% faster than Sonnet 5 and up to 30% lower cost per completed task for most work, even though the published input and output token prices are unchanged. That makes accepted output per task a more useful evaluation target than token price alone.
Source: Anthropic: Introducing Claude Sonnet 5.5.
What Claude Sonnet 5.5 is built to do
Sonnet 5.5 is aimed at the large middle of production work: tasks that need reliable execution and polish, but do not always require the slower, more expensive reasoning profile of Opus.
Coding and codebase work
The largest published jump is in agentic coding. Anthropic also reports fewer steps and more batched tool calls than Sonnet 5, which can shorten an iterative build-and-review loop.
Documents and analytical work
Sonnet 5.5 is positioned for polished documents, slides, spreadsheets, support work, research synthesis, and other well-scoped professional deliverables.
Design-sensitive output
Early testing highlighted stronger visual judgment for user interfaces and presentation templates. Treat this as a reason to evaluate real artifacts, not as a guarantee of design quality.
Fast collaboration
More than 30% faster output makes the model easier to steer in short loops. Higher effort can improve difficult work, but also increases time and cost per task.
Claude Sonnet 5.5 benchmark results
The table below summarizes Anthropic's published scores. The evaluations use different effort settings and conditions, so the rows should not be averaged into one universal quality number.
Swipe horizontally to compare every column.
| Evaluation | Published result | What it tests |
|---|---|---|
| Terminal-Bench 4.0 | 70.6% | Agentic terminal coding |
| FrontierCode 1.1 | 52.1% Xhigh | Merge-ready code changes |
| CursorBench 4.0 | 55.5% | Ambiguous multi-file coding tasks |
| GDPval-AA v2.1 | 1,844 | Professional work across occupations |
| AA-Briefcase v1.1 | 1,811 | Long-horizon knowledge work |
| Humanity's Last Exam | 64.5% with tools | Multidisciplinary reasoning |
| OSWorld 2.1 | 80.1% partial | Computer-use evaluation |
| Chartography | 61.6% no tools | Visual chart recognition |
Terminal-Bench 4.0 is the clearest generational jump: Anthropic reports 70.6% for Sonnet 5.5 versus 10.3% for Sonnet 5. On FrontierCode and CursorBench, the stronger result also depends on the selected effort level. Reproduce the task shape, tool setup, and scoring criteria before treating a public score as a budget forecast.
Sonnet 5.5 vs Sonnet 5 and Opus 5.5
Swipe horizontally to compare every column.
| Model | Best fit | Trade-off |
|---|---|---|
| Claude Sonnet 5.5 | Fast coding, everyday professional work, and high-volume iteration | Use effort settings and acceptance tests to control cost |
| Claude Sonnet 5 | Existing integrations that have not migrated yet | Slower and less capable on the newly published evaluations |
| Claude Opus 5.5 | Complex, open-ended work requiring sustained judgment | Higher published input and output token prices |
Sonnet 5.5 can approach Opus 5.5 on several structured evaluations, especially at higher effort. Anthropic still describes Opus 5.5 as clearly stronger for complex, open-ended work requiring sustained judgment. Route tasks by difficulty rather than assuming one model should handle every request.
Claude Sonnet 5.5 API pricing
Input tokens
$2 / 1M tokens
Output tokens
$10 / 1M tokens
Cache writes
$2.50 / 1M tokens
Cache reads
$0.20 / 1M tokens
These are Anthropic's published platform rates, not a quote for every intermediary route. Search the live LAYSO model catalog to confirm availability and the configured selling price. Measure cost per accepted task, including retries, tool calls, and review time.
API model ID and migration checklist
Anthropic documents the model ID as claude-sonnet-5-5. If the route is active on LAYSO, use the same account that manages your API key, balance, access, and request history.
curl https://api.layso.ai/v1/messages \
-H "x-api-key: $LAYSO_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-sonnet-5-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Review this release checklist."}]
}'- Verify route availability and the live selling price before a billable run.
- Re-test tool schemas, streaming, timeouts, retries, and preserved-thinking behavior.
- Compare low, medium, high, and Xhigh effort only on tasks that need them.
- Keep Sonnet 5 or another known route as a controlled rollback option.
- Review the new between_tools setting if your integration previously ran with thinking off.
Safety, safeguards, and production limits
Anthropic reports that Sonnet 5.5 improves on or matches Sonnet 5 across most automated alignment measures. Because its cyber capabilities are stronger, it launches with additional cyber safeguards and fallbacks; its biology safeguards remain aligned with Sonnet 5. Routine software development and most life-science work are not the target of those narrow high-risk controls.
Production teams still need their own authorization boundaries, human review, audit logs, data-retention decisions, and failure handling. A model safety report does not replace application-level controls.
Source: Claude Sonnet 5.5 System Card.
Claude Sonnet 5.5 FAQ
When was Claude Sonnet 5.5 released?
Anthropic announced Claude Sonnet 5.5 on September 28, 2026. It is the second model in the Claude 5.5 family.
What is the Claude Sonnet 5.5 API model ID?
Anthropic documents claude-sonnet-5-5 as the model ID. Confirm that the route is enabled in your provider or LAYSO catalog before sending production traffic.
How much does Claude Sonnet 5.5 cost?
Anthropic lists $2 per million input tokens, $10 per million output tokens, $2.50 per million cache-write tokens, and $0.20 per million cache-read tokens. LAYSO selling prices may differ, so check the live catalog.
Is Claude Sonnet 5.5 faster than Sonnet 5?
Anthropic reports output generation that is more than 30% faster and up to 30% lower cost per task for most work, while keeping the same published input and output token prices as Sonnet 5.
Should I use Sonnet 5.5 or Opus 5.5?
Sonnet 5.5 is positioned for well-scoped everyday tasks, coding fixes, documents, slides, spreadsheets, and fast iteration. Anthropic positions Opus 5.5 for complex, open-ended work that requires sustained judgment.
Does LAYSO independently verify the published benchmarks?
No. The benchmark values on this guide are clearly attributed to Anthropic. Run representative acceptance tests with your own prompts, tools, latency targets, and cost limits before deployment.