Claude Sonnet 5 arrives built for agents

Anthropic has released Claude Sonnet 5, the new mid-tier model in its lineup, and the pitch is squarely about agents. The company calls it its most agentic Sonnet yet, meaning it is built to make plans, call tools like browsers and terminals, and work through multi-step tasks on its own rather than waiting for the next instruction. The gains over Sonnet 4.6 show up in coding, tool use, and knowledge work, with better scores on BrowseComp, which tests agentic web search, and OSWorld-Verified, which tests actually driving a computer. One behavior Anthropic highlights: the model checks its own output without being asked to.

Pricing is the part worth noting. Through August 31 the launch rate is $2 per million input tokens and $10 per million output, after which it settles at the standard $3 and $15. Sonnet 5 is available now on every plan from Free to Enterprise, through Claude Code, the API, and the Claude platform, under the name claude-sonnet-5. Anthropic also says it refuses malicious requests more reliably than 4.6, and that its cybersecurity abilities are deliberately weaker than the Opus models, with cyber safeguards turned on by default. You can read the full announcement here.

Why it matters

If you build agents or write code with Claude, the cheaper and more capable mid tier is the model most teams will actually run in production, so the launch pricing through August is a real reason to benchmark it on your own workloads now.