Claude Sonnet 4.6 is a large language model developed by Anthropic, released in February 2026 as a major update to the Sonnet model family. It is designed to deliver frontier-level intelligence—particularly in coding, agentic planning, and computer use—at a more accessible price point than the Opus series. The model introduces adaptive thinking (also referred to as adaptive reasoning), a framework that allows the system to dynamically adjust its internal processing depth based on the complexity of the prompt.
Key to this version is the inclusion of effort levels, ranging from low to max effort. When set to max effort, the model utilizes its full reasoning capacity to tackle high-stakes tasks, such as deep architectural code review or complex financial modeling, that require extensive internal verification and self-correction. This is supported by a 1 million token context window (available in beta), which allows the model to reason effectively across massive codebases and long-form document sets in a single session.
Technical enhancements in Sonnet 4.6 include context compaction, a feature that automatically summarizes older parts of a conversation to maintain an effective context length during long-running agentic sessions. It also features improved tool integration, including web search and fetch tools that can autonomously write and execute code to filter and process data. Performance benchmarks indicate that Sonnet 4.6 achieves near-parity with flagship models on evaluations like SWE-bench Verified and OSWorld-Verified, making it a primary choice for autonomous software engineering and enterprise automation.