Anthropic ships Claude Sonnet 5 at $3/M input, closing to within 8% of Opus 4.8 on BrowseComp and OSWorld-Verified

Anthropic released Claude Sonnet 5 on the Claude API and all subscription tiers. The model is the new default for Free and Pro plans. It introduces improved planning, browser and terminal tool use, and autonomous multi-step execution previously limited to Opus-class models. Introductory pricing is set at $2 per million input tokens and $10 per million output tokens until August 31, 2026, then rises to $3/$15.

System-card evaluations show Sonnet 5 reduces undesirable behaviors versus Sonnet 4.6 and records lower cybersecurity capability than current Opus models. Early-access partners at Lovable and internal teams report completion of full software-engineering pull requests and end-to-end Salesforce workflows without intermediate prompting. OSWorld-Verified and BrowseComp curves demonstrate strict improvement over Sonnet 4.6 at every effort level while remaining below Opus 4.8 at maximum effort.

The release compresses the capability gap that previously forced developers to choose between Sonnet cost and Opus reliability for agentic workloads. Pricing and safety profile together lower the barrier for production agent deployment in coding and knowledge-work settings. Operational effect is immediate substitution of Sonnet 4.6 instances and selective replacement of lower-effort Opus calls.

Next measurable signal will be API traffic share and third-party agent-framework benchmarks within 90 days.