Claude Sonnet 4.6 Explained: Anthropic's Latest Mid-Tier AI Model

Release and Positioning

Claude Sonnet 4.6 is a large language model released by Anthropic on February 17, 2026, as the latest entry in the company's mid-tier "Sonnet" line. Anthropic has described it as a full upgrade of the Sonnet line's skills across coding, computer use, long-context reasoning, agent planning, knowledge work, and design, positioning it as the most capable Sonnet-class model the company has shipped to date.

Why a "Mid-Tier" Model Matters

Anthropic organizes its Claude lineup into roughly three tiers, often described informally as fast-and-light, balanced, and maximum-capability models. Sonnet sits in the middle tier, intended to deliver strong performance at a more practical cost than the company's top-tier "Opus" models. With Sonnet 4.6, Anthropic has said that performance which previously required an Opus-class model — including on real-world, economically valuable office tasks — is now available at Sonnet-tier pricing, narrowing the practical gap between its mid-tier and flagship offerings.

Technical Highlights

Sonnet 4.6 ships with a one-million-token context window available in beta, allowing it to process very large inputs such as entire codebases or lengthy collections of documents within a single request. On the Message Batches API, it supports output of up to 300,000 tokens using a dedicated beta header. Its model identifier on the Claude API is claude-sonnet-4-6, and unlike many earlier Claude model IDs that included a specific release date, Sonnet 4.6 uses a dateless naming format that is nonetheless still a fixed, pinned model snapshot rather than an evergreen, automatically updating model.

Computer Use and Agentic Tasks

A particular focus of the Sonnet 4.6 release was computer use — the ability for the model to operate a computer interface directly, similar to how a human would click, type, and navigate between applications. Anthropic reported that early users observed close to human-level capability on tasks like navigating a complex spreadsheet or filling out a multi-step web form, and that the model showed meaningful improvement in resisting prompt injection attacks, in which malicious actors try to hijack an AI agent by hiding instructions inside web content the agent processes.

Pricing and Availability

Despite the capability improvements, Anthropic kept Sonnet 4.6's pricing the same as its predecessor, Sonnet 4.5, at $3 per million input tokens and $15 per million output tokens through the API, with additional savings available through prompt caching and batch processing. For users of Claude's free and Pro consumer plans, Sonnet 4.6 became the default model in the Claude.ai interface and in Claude Cowork following its release.

Safety Evaluation

As with every new Claude release, Anthropic conducted extensive internal safety evaluations of Sonnet 4.6 before launch. The company's safety researchers concluded that the model displayed a broadly warm, honest, and prosocial character, strong safety behaviors, and no signs of major concerns around high-stakes misalignment, in line with the testing standards Anthropic applies across its model lineup.