AI

Anthropic Releases Claude Sonnet 4.6 with Enhanced Coding and Computer Use Capabilities

Anthropic has announced the release of Claude Sonnet 4.6, marking a significant upgrade to its Sonnet family of AI models. The new model introduces improvements across multiple domains including coding, computer use, long-context reasoning, agent planning, knowledge work, and design, while offering a one million token context window now available in beta.

The model has become the default option for users on Anthropic’s Free and Pro plans across claude.ai and Claude Cowork platforms. Pricing remains unchanged from the previous version, starting at $3 per million tokens for input and $15 per million tokens for output.

“We’re excited to bring these capabilities to more users,” the company stated in its announcement. The release represents what Anthropic describes as the most capable Sonnet model to date, with particular emphasis on developer-oriented features.

Coding Performance Gains

Early access users with development backgrounds have demonstrated a strong preference for Claude Sonnet 4.6 over its predecessor, Sonnet 4.5. According to the company, these developers often prefer the new model even compared to Claude Opus 4.5, which was released in November 2025 and represented the company’s most advanced offering at that time.

The improvements are particularly notable in consistency, instruction following, and overall coding task completion. The company reports that performance levels previously requiring Opus-class models are now accessible through Sonnet 4.6, including economically valuable office tasks that benefit from advanced AI capabilities.

Early testing in Claude Code found that users preferred Sonnet 4.6 over Sonnet 4.5 approximately 70% of the time. Testers reported that the model more effectively reads context before modifying code and consolidates shared logic rather than duplicating it, reducing frustration during extended usage sessions.

Customers have also noted improvements in frontend code generation and financial analysis. Visual outputs from the model have been described as notably more polished, with better layouts, animations, and design sensibility compared to previous versions. Organizations reported requiring fewer iteration rounds to achieve production-quality results.

Computer Use Capabilities Advance

Claude Sonnet 4.6 demonstrates substantial improvement in computer use skills compared to prior Sonnet models. The company noted that while AI models historically required special APIs or purpose-built connectors to interact with software, the new model can operate computers in a manner similar to human users.

The OSWorld benchmark, which evaluates AI computer use across hundreds of tasks using real software including Chrome, LibreOffice, and VS Code running on simulated computers, shows significant progress over sixteen months of development. In these tests, the model sees and interacts with the computer through virtual mouse clicks and keyboard input, without specialized interfaces.

Early users have reported human-level capability in tasks such as navigating complex spreadsheets and completing multi-step web forms, subsequently integrating actions across multiple browser tabs.

However, the company acknowledges that the model still lags behind the most skilled humans in computer use. Despite this, the rate of progress has been described as remarkable, expanding the practical utility of computer use for a range of work tasks.

Safety and Security Considerations

As with each new Claude model release, Anthropic conducted extensive safety evaluations of Sonnet 4.6. The overall findings indicate the model is as safe as, or safer than, other recent Claude models. The company’s safety researchers characterized Sonnet 4.6 as having “a broadly warm, honest, prosocial, and at times funny character, very strong safety behaviors, and no signs of major concerns around high-stakes forms of misalignment.”

The company also addressed security concerns related to computer use, specifically prompt injection attacks where malicious actors attempt to hide instructions on websites to hijack model behavior. Safety evaluations demonstrate that Sonnet 4.6 shows major improvement in resistance to prompt injections compared to Sonnet 4.5, performing similarly to Opus 4.6 in this regard.

Benchmark Performance and Context Window

Claude Sonnet 4.6 has improved across benchmarks, approaching Opus-level intelligence at a more accessible price point. The model matches Opus 4.6 performance on OfficeQA, which measures how well a model can read enterprise documents including charts, PDFs, and tables, extract relevant information, and reason from that information.

The one million token context window allows the model to hold entire codebases, lengthy contracts, or dozens of research papers in a single request. More importantly, the model reasons effectively across all this context, demonstrating improved capability for long-horizon planning.

In the Vending-Bench Arena evaluation, which tests how well a model can run a simulated business over time with competitive elements where different AI models compete to generate the largest profits, Sonnet 4.6 developed an innovative strategy. The model invested heavily in capacity during the first ten simulated months, spending significantly more than competitors, then pivoted sharply to focus on profitability in the final stretch. This strategic timing helped the model finish ahead of the competition.

Industry Response

Early customers across the technology sector have provided positive feedback on the model’s capabilities.

Hanlin Tang, CTO of Neural Networks at Databricks, commented: “The performance-to-cost ratio of Claude Sonnet 4.6 is extraordinary. Sonnet 4.6 outperforms on our orchestration evaluations, handles our most complex agentic workloads, and keeps improving as effort settings increase.”

Michele Catasta, President of Replit, noted: “Claude Sonnet 4.6 is a notable improvement over Sonnet 4.5 across the board, including long-horizon tasks and more difficult problems.”

Michael Truell, Co-founder and CEO of Cursor, observed: “Out of the gate, Claude Sonnet 4.6 is already excelling at complex code fixes, especially when searching across large codebases is essential. For teams running agentic coding at scale, we’re seeing strong resolution rates and the kind of consistency developers need.”

Joe Binder, VP of Product at GitHub, stated: “Claude Sonnet 4.6 has meaningfully closed the gap with Opus on bug detection, letting us run more reviewers in parallel, catch a wider variety of bugs, and do it all without increasing cost.”

Scott Wu, CEO of Cognition, offered: “For the first time, Sonnet brings frontier-level reasoning in a smaller and more cost-effective form factor. It provides a viable alternative if you are a heavy Opus user.”

The release of Claude Sonnet 4.6 represents Anthropic’s continued development of AI models targeting enterprise and developer use cases, with particular emphasis on balancing capability improvements with practical considerations around cost and accessibility.


Source: https://www.anthropic.com/news/claude-sonnet-4-6