Claude Sonnet 4.6

Claude Sonnet 4.6 is the default model for Free and Pro plans on claude.ai and Claude Cowork, delivering a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. It features a 1-million-token context window (in beta) and represents a comprehensive step up from Sonnet 4.5 in every measured dimension. In Claude Code, users preferred Sonnet 4.6 over Sonnet 4.5 roughly 70% of the time, and preferred it over the previous-generation Opus 4.5 59% of the time — citing less overengineering, less "laziness," better instruction following, fewer false claims of success, and fewer hallucinations.

Computer use capabilities have advanced dramatically. OSWorld scores doubled in 16 months, reaching human-level capability on complex spreadsheets, multi-step web forms, and multi-tab workflows. On an insurance benchmark for submission intake and first notice of loss, Sonnet 4.6 scores 94% — the highest-performing model for computer use in that domain. Prompt injection resistance has seen a major improvement over Sonnet 4.5, now on par with Opus 4.6, making it significantly more secure for autonomous computer use in production environments.

On business and enterprise benchmarks, Sonnet 4.6 matches Opus 4.6 on OfficeQA (enterprise document comprehension) and shows a 15 percentage point improvement over Sonnet 4.5 on Box's heavy reasoning Q&A evaluation. On Vending-Bench Arena, the model invested heavily in capacity for the first 10 simulated months, then pivoted to profitability — almost tripling Sonnet 4.5's average gains over a full year of simulated business operation. Frontend code and financial analysis are standout improvements, with visual outputs described as "notably more polished" featuring better layouts, animations, and design sensibility, requiring fewer rounds of iteration to reach production quality.

The model supports both adaptive thinking and extended thinking for flexible reasoning depth. Context compaction is available in beta, automatically summarizing older context when approaching the token threshold. Web search and fetch tools now auto-write and execute code to filter results, making research workflows more autonomous. Code execution, memory, programmatic tool calling, tool search, and tool use examples are all now generally available.

Claude in Excel now supports MCP connectors for major financial data providers including S&P Global, LSEG, Daloopa, PitchBook, Moody's, and FactSet — enabling direct data pulls into spreadsheet workflows without leaving the Excel environment. These integrations are free for all users through MCP-based connectors, alongside Google Workspace and Slack integrations.

At $4/$20 per million tokens through CallMissed (or $3/$15 directly from Anthropic), Sonnet 4.6 is the recommended model for most production workloads. Its combination of near-Opus quality on reasoning and document comprehension, dramatically improved computer use, top-tier financial analysis performance, enhanced prompt injection resistance, and Sonnet-level pricing makes it the best value proposition in the Claude family for code generation, enterprise document analysis, computer use automation, and general-purpose assistance.