A Sonnet model designed for agents
Anthropic released Claude Sonnet 5 on June 30, describing it as its most agentic Sonnet model yet. The company says it improves reasoning, tool use, coding and knowledge work over Sonnet 4.6 while approaching the performance of more expensive Opus-class models on some tasks.
Availability is immediate across Claude plans, Claude Code and the Claude Platform. That makes the release especially relevant to developers: the model can be tested inside an existing coding-agent workflow instead of only through isolated prompts or a benchmark table.
Pricing needs a date attached
Anthropic launched Sonnet 5 with introductory API pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026. The stated standard pricing after that period is $3 per million input tokens and $15 per million output tokens.
Any cost comparison published during the introductory window should show both schedules. Coding agents can consume large contexts and produce long traces, so token price is only one part of the bill. Cache behavior, effort settings, retries and the time needed to review a patch can materially change the result.
Test the harness, not only the model
A reliable Claude Code evaluation should use a real repository and a bounded issue with an objective test. Record whether the agent finds the right files, asks for missing context, keeps its changes in scope and explains failures. Then review security-sensitive operations and whether the patch is easy to revert.
Sonnet 5 may raise the capability ceiling, but model intelligence does not replace repository instructions, permission boundaries or human code review. The practical gain comes when stronger planning reduces interventions without increasing hidden errors.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.