Anthropic has released Claude Sonnet 4.6, the latest version of its Sonnet model family, expanding capabilities across coding, computer use, reasoning, and design. The model is now the default option for users on Free and Pro plans in claude.ai and Claude Cowork, replacing Sonnet 4.5 at the same price point.
Claude Sonnet 4.6 is priced at $3 per million input tokens and $15 per million output tokens, unchanged from the previous version. The model introduces a 1 million token context window in beta, allowing it to process entire codebases, lengthy contracts, or large collections of documents in a single request.
Anthropic said the release reflects a full upgrade rather than incremental tuning. Early access users reported strong gains in instruction following, consistency, and long-session reliability, particularly for software development tasks. In internal testing, developers preferred Sonnet 4.6 over Sonnet 4.5 roughly 70% of the time, and favored it over the higher-end Claude Opus 4.5 model in a majority of cases.
Improved Coding and Computer Use
A major focus of Sonnet 4.6 is computer use, a capability Anthropic first introduced in late 2024. Computer-using models interact with software visually, clicking and typing like a human rather than relying on APIs. This approach enables automation in legacy or specialized systems that lack modern interfaces.
Performance on OSWorld, a benchmark that measures AI interaction with real software such as browsers, spreadsheets, and code editors, shows steady improvement across recent Sonnet releases. Anthropic said Sonnet 4.6 demonstrates near human-level performance on tasks such as navigating complex spreadsheets or completing multi-step web workflows, though it still trails expert human users.
Anthropic also said safety evaluations show improved resistance to prompt injection attacks, a risk where malicious instructions are hidden inside web pages. According to the company, Sonnet 4.6 performs significantly better than Sonnet 4.5 and aligns closely with Opus 4.6 on safety benchmarks.
Long-Context Reasoning at Lower Cost
Beyond computer use, Sonnet 4.6 shows gains in long-horizon reasoning and planning. Anthropic highlighted improved performance on benchmarks that test multi-step decision-making over extended contexts, including simulations where models manage resources over time.
The expanded context window, combined with new context compaction tools in beta, allows the model to summarize older parts of a conversation automatically, extending effective working memory without increasing token usage.
On the Claude Developer Platform, Sonnet 4.6 supports adaptive and extended thinking modes, along with broader access to tool use, code execution, and web search features. Anthropic said these updates are aimed at making advanced AI capabilities more accessible without requiring Opus-class pricing.
Sonnet 4.6 is available now across all Claude plans, the Claude API, Claude Code, and major cloud platforms. Anthropic said Opus 4.6 will remain its strongest option for tasks requiring the deepest reasoning, but Sonnet 4.6 narrows the gap for a wide range of real-world work.
