TL;DR: Anthropic has released Claude Sonnet 4.6, which it describes as delivering near-Opus-level performance at Sonnet pricing ($3/$15 per million tokens). The model features significant improvements in coding, computer use, and agent planning, with a 1M token context window in beta.

Closing the Gap with Opus

In internal testing, Claude Code users preferred Sonnet 4.6 over its predecessor roughly 70% of the time, citing better code comprehension and less duplication. Users even preferred it to Opus 4.5 — Anthropic’s most powerful model from November 2025 — 59% of the time, rating it less prone to overengineering and more consistent at following instructions.

The model matches Opus 4.6 on OfficeQA, a benchmark measuring how well AI can read enterprise documents including charts, PDFs and tables. Early customers highlighted strong improvements in frontend code quality and financial analysis.

Computer Use Advances

Sonnet 4.6 marks continued progress in Anthropic’s computer use capability, first introduced in October 2024. On OSWorld — a benchmark testing AI on real software tasks across Chrome, LibreOffice and VS Code — Sonnet models have shown steady gains over sixteen months.

Early users report “human-level capability” in tasks like navigating complex spreadsheets, filling multi-step web forms and working across multiple browser tabs. Anthropic notes the model has improved resistance to prompt injection attacks compared to Sonnet 4.5.

Platform Updates

The release includes several developer-focused features: adaptive and extended thinking, context compaction (which automatically summarises older context as conversations approach limits), and enhanced web search tools that can write and execute code to filter results. For Claude in Excel users, the add-in now supports MCP connectors for tools like S&P Global, LSEG and PitchBook.

Looking Forward

Sonnet 4.6’s performance-to-cost ratio is particularly relevant for UK businesses running AI at scale. Tasks that previously required Opus-class models — and Opus-class pricing — can now be handled by Sonnet 4.6 in many cases. Anthropic still positions Opus 4.6 as the strongest option for deep reasoning tasks like codebase refactoring, but the gap is narrowing with each Sonnet release.