Claude Sonnet 5 brings Opus-class agents down to mid-tier prices

Anthropic released Claude Sonnet 5 on June 30, and the significant part is not another benchmark race — it is the economics of running AI agents. The company positions Sonnet 5 as its most agentic mid-tier model yet: it plans, drives browsers and terminals, and completes multi-step work that until recently required Opus-class models. On agentic evaluations it lands close to Opus 4.8, but at $3 per million input tokens and $15 per million output (with introductory pricing of $2/$10 through August 31) versus Opus 4.8 at $5/$25. A configurable "effort" level lets teams trade cost against capability on the same model, so one deployment can span cheap routine automation and harder reasoning tasks. Two operational details deserve attention before migrating workloads: a new tokenizer maps the same input to roughly 1.0–1.35× more tokens (Anthropic says intro pricing keeps the transition roughly cost-neutral), and real-time cyber safeguards are now enabled by default. The model is available on the Claude Platform, AWS, and Microsoft Foundry, with Google Vertex support coming. For teams building agentic workflows, the practical takeaway is that the price floor for capable, autonomous multi-step agents just dropped by roughly half — worth re-running your model-selection math.

This is a summary by our content curator. Read the original at Anthropic: https://www.anthropic.com/news/claude-sonnet-5.