Anthropic announced Claude Sonnet 5, the latest version of its mid-tier model, on June 30, 2026, describing it as "our most agentic Sonnet yet." The company made the model available the same day to all users across its Free, Pro, Max, Team and Enterprise plans, as well as to developers through Claude Code and the Claude Platform. According to Anthropic, Sonnet 5 can make plans, use tools such as browsers and terminals, and operate autonomously at a level that only a few months ago required larger and more expensive models.
The model is offered at an introductory price of $2 per million input tokens and $10 per million output tokens. That rate is scheduled to rise to $3 per million input tokens and $15 per million output tokens after August 31, 2026. Anthropic said Sonnet 5 shows clear improvements in reasoning, tool use, software coding and knowledge work over its predecessor, Sonnet 4.6, which was released in February 2026. The coding tool Cursor reported that Sonnet 5 scored 57% on its CursorBench benchmark, up from 49% for Sonnet 4.6.
Anthropic emphasized safety and reliability in the release, particularly for agentic use. The company said Sonnet 5 exhibits a lower rate of undesirable behaviors, such as cooperation with misuse and deception, than earlier Sonnet versions, and is better at refusing malicious requests and resisting prompt-injection attacks. The model ships with real-time cyber safeguards similar to those used in Opus-class models. Early access partners said it performs reliably on complex technical tasks and follows through on multi-step assignments.
The launch lands as mid-tier models increasingly reshape the price-performance balance in the AI market. Anthropic's benchmark data indicate that Sonnet 5 narrows the gap with the higher-end Opus 4.8 at a significantly lower cost, making it attractive for organizations seeking a balance between price and advanced capability. On the same day, China's Meituan released its 1.6-trillion-parameter LongCat-2.0 model, underscoring that competition among open and closed model developers has not slowed over the summer.