Claude Sonnet 4.6 is a mid-tier large language model released by Anthropic in February 2026. Positioned as a balance between speed and intelligence, it is designed to deliver performance comparable to frontier-class models like Claude Opus while maintaining the cost-efficiency of the Sonnet series. The model features a significant upgrade in agentic planning, coding, and computer use, notably achieving a 72.5% score on the OSWorld-Verified benchmark for autonomous computer interaction.
The Low Effort (or Non-reasoning) configuration specifically refers to the model's operation when its Extended Thinking capabilities are disabled or minimized. While Sonnet 4.6 is capable of internal chain-of-thought reasoning through its Adaptive Thinking feature, the low-effort mode is optimized for maximum speed and lower token latency. In this state, the model provides direct responses without the overhead of an internal reasoning block, making it ideal for routine instructions, high-volume data extraction, and low-complexity chat applications.
Technical Capabilities
Sonnet 4.6 supports a massive 1 million token context window in beta, allowing it to ingest entire codebases, long legal documents, or extensive research archives. To manage this context effectively, Anthropic introduced context compaction, which automatically summarizes older parts of a conversation as it approaches limits to preserve essential information. This enables longer, more coherent agentic workflows and prevents performance degradation in multi-hour tasks.
For production environments, the model is highly resistant to prompt injections and features a more prosocial, conversational tone compared to previous generations. Official guidance suggests utilizing the low-effort setting for tasks where the output is deterministic or the instructions are straightforward, reserving the high-effort reasoning modes for complex debugging, mathematical proofs, and long-horizon planning.