Claude Sonnet in one minute
Examples include rewriting a draft, analysing documents, changing code or running an agent with clear intermediate steps. Sonnet is a candidate when you want to respond regularly to work in progress. Opus or Fable may suit more complex assignments, but assess that with examples from your own work.
Sonnet’s current position
Anthropic introduced Sonnet 5.5 on 28 September 2026. Current API documentation places its speed in the “fast” category. The family targets broadly useful work, including programming and agents. This makes Sonnet a practical candidate for processes involving several rounds of user feedback.
A quick first response is only one part of productivity. If a model needs frequent restarts, the entire task may take longer. Compare time to a usable result. Current standard API token prices place Sonnet below Opus and Fable, but tools, reasoning and repeated attempts affect the total bill.
Sonnet does not permanently mean “the middle model”. The original Claude 3 family centred on Opus, Sonnet and Haiku. Fable and Mythos now also exist, and families receive updates at different times. The model catalogue is the source for current relationships.
Writing, analysis and everyday work
For editorial work, Sonnet can restructure rough text, compare alternatives and apply a consistent tone. Supply the facts separately. “Make this more persuasive” may encourage additional claims; “improve the structure without adding product features” provides a clearer criterion.
For analysis, ask about similarities and differences between documents, or questions a report leaves unanswered. An account manager might compare three versions of a brief and then check the changed requirements. The value lies in a clearer overview, not automatic approval of every conclusion.
For code, Sonnet is a candidate for everyday changes, explanations and debugging. Provide the relevant project context and, where possible, execute and check changes in an appropriate environment. A proposed solution becomes convincing when the observed behaviour and verification support it.
Sonnet inside an agent
An agent needs tools, instructions and a working environment alongside its model. Sonnet can help choose the next step, interpret a tool result and adapt the approach. The application determines whether it may actually change a file or operate another system.
For example, an assistant could organise interview notes and produce a draft brief. Define completion clearly: every source processed, contradictions identified and open questions listed separately. This makes the process assessable without treating response length as a quality measure.
Context and reasoning settings
Sonnet 5.5 uses the API model ID claude-sonnet-5-5. Documentation lists 1 million tokens of context, text and image input, and text output. This does not imply that every app offers identical input limits or uploads. A large context window makes extensive material available but does not guarantee complete or error-free processing.
This version supports adaptive thinking. A reasoning setting controls how much processing effort the model devotes to a task; it does not turn Sonnet into Opus or Fable. Defaults can also differ by surface: API documentation specifies high, while the launch information describes medium for the apps and Claude Code. Check the actual environment before comparing results.
Measure speed and cost with the settings you will use. A trial with short answers and no tools says little about an agent that reads multiple files and executes commands.
From earlier generations to Sonnet 5.5
Sonnet became part of the Claude 3 family in 2024. Later steps included 3.5, 3.7, 4, 4.5, 4.6 and 5, followed by the current 5.5. This is development history; it does not mean every version remains available.
Lifecycle documentation records retirement of Sonnet 3.5 on 28 October 2025, Sonnet 3.7 on 19 February 2026 and Sonnet 4 on 15 June 2026. Sonnet 4.5 has been deprecated since 30 September 2026, with retirement scheduled for 30 November 2026. On the review date, that last version therefore still operates under the announced API schedule.
Sonnet 4.6 and 5 remain listed as active. The family as a whole is not a legacy family. This lifecycle information applies to the specified API channels; partner platforms and the Claude app can have their own availability.
Sonnet or Opus?
Choose a real task, such as rewriting a landing page with ten fixed facts or making a small repository change. Assess whether Sonnet follows the requirements, what corrections are needed and how long the complete process takes. Test Opus when the assignment contains considerable ambiguity, dependencies or trade-offs.
Fable may be relevant for especially sustained and difficult analysis. Haiku is another candidate for fast, bounded processing. Mythos requires specific verification and is not an ordinary next subscription tier. The appropriate choice follows from sufficient quality at a suitable time and cost.