Anthropic has launched Claude Sonnet 4.6, a new version of its Sonnet model, now available across all Claude plans, including Claude Cowork, Claude Code, the API, and major cloud platforms.
The company has also upgraded its free tier to offer Sonnet 4.6 as the default model, which includes features such as file creation, connectors, skills, and context compaction.
According to Anthropic, Sonnet 4.6 introduces a full refresh in areas, including coding capability, computer use, long-context reasoning, agent planning, knowledge work, and design.
The model features a one-million-token context window in beta, enabling it to process larger volumes of information, such as entire codebases or lengthy contracts, within a single request.
Pricing for Sonnet 4.6 remains unchanged from the previous version at $3 or $15 per million tokens.
Early users have reported that Sonnet 4.6 demonstrates significant improvements over prior iterations when performing tasks such as navigating complex spreadsheets or completing multi-step web forms.
In addition, developers with early access indicated a preference for Sonnet 4.6 compared to both its immediate predecessor and Opus 4.5 for tasks involving code consistency and instruction following.
The company states that the new model shows increased reliability in avoiding errors such as false claims of success or hallucinations during multi-step processes.
Benchmark evaluations like OSWorld have tracked steady progress in the Sonnet models’ ability to interact with real software through simulated environments without requiring special connectors or APIs.
In business simulation tests like Vending-Bench Arena, designed to measure AI performance in running enterprises over time, Sonnet 4.6 applied strategies that resulted in higher outcomes compared to competitors by shifting focus from capacity investment to profitability at critical stages.
Further updates include expanded support on the Claude Developer Platform for adaptive and extended thinking alongside context compaction (in beta), which summarises older context automatically as conversation length approaches system limits.
On the API side, web search and fetch tools can now autonomously write and execute code to enhance filtering and processing of search results; this is offered together with tools for code execution, memory management, programmatic tool calling, and more.
Anthropic reports that safety assessments found Sonnet 4.6’s performance comparable or superior to earlier models in resisting risks such as prompt injection attacks that could influence behaviour via hidden instructions on websites.
Sonnet 4.6 aims to make advanced AI capabilities accessible at a lower price point while maintaining security standards seen in previous Anthropic releases.
