[Anthropic Launches Claude Sonnet 5, Significant Improvement in Agent Capabilities] Jin10 Data, July 1 — Anthropic has launched the Claude Sonnet 5 model. Anthropic stated that this model is the most capable Sonnet model to date in terms of agent capabilities, capable of planning, utilizing tools like browsers and terminals, and autonomously operating at a level that previously required larger and more expensive models just months ago. Sonnet 5's performance approaches that of Opus 4.8 while being more cost-effective, with significant enhancements over the previous Sonnet 4.6 in key agent dimensions such as reasoning, tool usage, programming, and knowledge work. Safety assessments show that Sonnet 5 has a lower incidence of undesirable behavior compared to Sonnet 4.6, making it safer in agent scenarios, though its cybersecurity task capabilities remain far below current Opus models. Starting today, Sonnet 5 is available across all subscription plans: Free and Pro versions use it by default, while Max, Team, and Enterprise users can opt for it. The model is also available on the Claude Code and Claude platforms. During the launch period, pricing is set at $2 per million input tokens and $10 per million output tokens, with the promotional period lasting until August 31, 2026, after which prices will adjust to $3/$15.