Claude Sonnet 5.5 API for Coding and Agentic Work

Claude Sonnet 5.5 is Anthropic's September 28, 2026 update to Sonnet 5. It accepts text and images and returns text. Anthropic positions it for coding, analysis, visual understanding, and agentic tool use; Opus 5.5 remains better suited to complex, open-ended work.

It keeps Sonnet 5's $2 input / $10 output per million tokens on Anthropic's direct API. Anthropic reports output generation more than 30% faster and up to 30% lower cost per task through fewer tokens, rather than a lower per-token rate. GPTProto's actual charge appears in the fixed pricing module.

Claude Sonnet 5.5 Applications

Repository changes and code review. Give an agent a defined issue, relevant files, tests, and acceptance criteria. Let the application supply tools and validate the result. Escalate difficult architecture decisions.

Long-document analysis. Supply specifications, support records, or reports and request structured answers with source references. The 1M-token context window adds capacity, but extraction checks still matter.

Image-aware workflows. Send screenshots or charts with text to explain a UI defect or extract information. This is image understanding with text output, not image generation. Verify GPTProto's image-input route before deployment.

Sonnet 5.5 vs Sonnet 5 vs Opus 5.5

These are Anthropic direct API prices and Anthropic-reported evaluation results. They are provided for model selection; the GPTProto pricing widget is the source for GPTProto billing.

The same per-token rate does not mean the same bill. Log tokens, cache usage, tool calls, retries, and accepted results. Benchmarks help orient model selection, but effort and task setup change total cost. Route repeatable steps to Sonnet 5.5 and evaluate Opus 5.5 for harder decisions.

Migrating from Claude Sonnet 5

Replacing the model string alone can break an integration. Check the ID and request format in GPTProto's API Usage tab. If the route exposes Anthropic's native controls, test these changes before moving live traffic:

1. Replace claude-sonnet-5 with the supported Sonnet 5.5 ID. If you used thinking: {"type": "disabled"}, use thinking: {"type": "between_tools"} at low, medium, or high effort; disabled returns a 400 error on the new model.

2. Remove forced tool_choice values any and tool. Use auto, validate tool input, and test whether the model calls a required tool in your workflow.

3. Parse response blocks by type and preserve thinking blocks when continuing a conversation. Review streaming UIs that previously displayed text between tool calls.

4. Recheck max_tokens, effort, and non-default sampling parameters. Thinking consumes output tokens, while custom temperature, top_p, or top_k values can return a 400 error.

Test the exact payload on your chosen GPTProto route; native Anthropic controls may differ across routes.

When Should You Choose Sonnet 5.5?

Start with Sonnet 5.5 for repeatable code fixes, document work, and agent steps with clear success criteria. Compare Claude Opus 5.5 for extended judgment. When migrating from Claude Sonnet 5, measure cost per accepted result. Explore the model catalog to test other providers with one account.