Claude Sonnet 4.6 is positioned by Anthropic as its best general-purpose model for price-performance, with the release framed around teams that build coding assistants and agentic software workflows. The model is a direct successor to Sonnet 4.5 and is designed for the kind of real engineering work that goes beyond self-contained snippets, including messy repositories, brittle tests, ambiguous tickets, and multi-step tool interaction, which is why SWE-bench Verified, Terminal-Bench 2.0, and OSWorld-Verified are highlighted as the most legible signals of its production value.

On those headline engineering evaluations, Sonnet 4.6 reports 79.6% on SWE-bench Verified, 59.1% on Terminal-Bench 2.0 under default thinking, and 72.5% on OSWorld-Verified, representing meaningful gains over Sonnet 4.5, with the OSWorld-Verified jump in particular pointing to stronger computer-use and desktop-agent capability. Pricing is reported as $3 per million input tokens and $15 per million output tokens, unchanged from the prior Sonnet tier, reinforcing its positioning as a high-throughput, cost-conscious model for coding and agentic tasks where the long-context, multimodal, and tool-calling characteristics of the Claude line remain central to the practical fit.