Claude Sonnet 5 is Anthropic's Sonnet model for programming, reasoning, and agentic work, with a focus on improving sustained execution from task planning to tool collaboration. It supports analyzing problems using both text and images, making it suitable for code maintenance, technical research, and knowledge organization. On this platform, you can choose to orchestrate messages yourself or use managed multi-turn conversations to complete continuous tasks.
Clarify capacity, input/output, and invocation methods before selecting a model.
Reasoning and visual understanding are model capabilities; file reading, session management, and tool execution are provided by the selected endpoint and authorization configuration.
Learn what claude-sonnet-5 can bring to your work.
The focus of Sonnet 5 is not just completing code, but planning continuous work around a goal. Compared with Sonnet 4.6, it offers improvements in programming, reasoning, and tool use. Have it first break down the scope of changes, then generate implementation suggestions and a test plan; after connecting execution tools, continue adjusting based on runtime feedback.
For tasks that require querying, reading, and verification, it is well suited to take on the decision-making role in an agent. Applications can provide function tools, allowing the model to advance work based on returned results; hosted conversations can also use authorized tools for multi-turn collaboration. Execution permissions and operable objects are still set by the application, rather than obtained by the model itself.
Text and images can jointly form the context of a problem, for example by placing error screenshots, interface states, and requirement descriptions in the same input turn. Sonnet 5 can provide explanations and troubleshooting suggestions based on this, making it suitable for turning visual information into discussable textual conclusions rather than mistaking image understanding for image generation capability.
Start with specific tasks to find where the model can be effective.
Provide relevant code, error logs, and expected behavior, and ask the model to output issue hypotheses, modification plans, and regression test suggestions. When testing tools are already available, return the results to continue narrowing down the issue. Deliverables should be clearly defined as patch suggestions, test cases, and items to verify, making them easier for engineers to review before merging.
Give the model technical documents, screenshots, and questions that need comparison to organize differences between approaches, implementation dependencies, and risks. When files need to be read, use the file content blocks and reading tools in the conversation interface. Finally, require the report to be organized by conclusions, evidence, and open questions so the team can continue making decisions.
Suitable for continuously organizing requirements, revising plans, and tracking issues around the same project. Use a hosted session ID to carry forward subsequent discussions, and connect authorized knowledge bases or code collaboration tools when needed. Limiting each turn's objective to a clear analysis or deliverable can reduce continuous expansion of task scope.
Choose based on task complexity, input materials, and expected results.
If existing tasks are often interrupted between planning, tool feedback, or code modifications, Sonnet 5 is worth prioritizing for testing. Its improvements focus on reasoning, programming, tool use, and knowledge work. When migrating, reuse real examples, compare accuracy and the amount of follow-up correction, and recalculate Token usage rather than copying the old budget.
Sonnet 5 is suitable for everyday engineering and multi-step analysis that require a balance between capability and investment; in official evaluations, higher reasoning effort can match Opus 4.8 on some tasks, but this does not mean it is equivalent across the board. For extremely difficult judgments, complex reviews, or critical deliverables, conduct parallel validation before deciding whether to use a higher-capability model.
From a small-scale task to production integration.
Define the goal, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to view the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before formal use, understand output quality and capability scope.
Answers to common questions about using claude-sonnet-5.
It is mainly suited for programming, research, and knowledge work that require continuous reasoning and tool feedback, such as breaking down modification plans and revising approaches based on test results. If you already have a stable Sonnet 4.6 workflow, you can first compare it using the same set of examples; there is no need to migrate immediately just because of the version change.
It can analyze code, propose changes, and participate in tool collaboration, but running a project requires an available execution environment and permissions. When only text is sent, you receive code or suggestions; after connecting tools, it can iterate based on execution feedback. Testing and merge reviews should still be retained in the end.
Combine text and image_url blocks in the message content of Chat Completions, clearly stating the issue shown in the screenshot and the question you want answered. Hosted conversations can also use a message array to organize text and images. Images are used for understanding and analysis, and responses are presented as text.
You can use the file_url content block of /aichat2/conversations together with the file-reading workflow for summarization or analysis. File reading is the processing method provided by the endpoint; do not submit a PDF address to Chat Completions as an image address, and important conclusions should be verified against the original text.
Hosted conversations can preserve sessions through stateful and continue the discussion in subsequent requests by bringing back the same id. When orchestrating Chat Completions yourself, the application maintains the required messages and tool results. The former is suitable for ongoing collaboration, while the latter makes it easier to control context organization.
Model information · Updated: 2026-10-01. For invocation parameters and billing rules, see the API and pricing sections.
Start with clear goals and determine from real results whether it is suitable for your work.
Input
$0.40per 1M tokens
Official $2.00
Output
$2.00per 1M tokens
Official $10.00
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs