News

Anthropic Releases Claude Sonnet 4.6 with Major Coding and Computer Use Improvements

· Anthropic

Anthropic has launched Claude Sonnet 4.6, marking a significant advancement in AI capabilities that could reshape the competitive landscape for coding assistants and computer automation tools. The new model delivers substantial improvements across coding, computer use, long-context reasoning, and agent planning while maintaining the same pricing as its predecessor at $3-15 per million tokens. The most notable enhancement comes in coding performance, where early users prefer Sonnet 4.6 over its predecessor by a wide margin. In testing with Claude Code, users chose the new model 70% of the time over Sonnet 4.5, citing better context reading, improved code consolidation, and reduced duplication. Perhaps more striking, users preferred Sonnet 4.6 to Claude Opus 4.5 - Anthropic's previous flagship model - 59% of the time, reporting less overengineering, fewer hallucinations, and better instruction following. Computer use capabilities represent another major leap forward. Since introducing the first general-purpose computer-using model in October 2024, Anthropic has steadily improved performance on OSWorld, the standard benchmark for AI computer interaction. Sonnet 4.6 demonstrates human-level capability in complex tasks like navigating spreadsheets and filling multi-step web forms across browser tabs. The model interacts with software through virtual mouse clicks and keyboard input, eliminating the need for custom APIs or connectors. The model features a 1-million token context window in beta, sufficient to process entire codebases, lengthy contracts, or dozens of research papers in a single request. This expanded context enables more effective long-horizon planning, demonstrated in the Vending-Bench Arena evaluation where Sonnet 4.6 developed a sophisticated business strategy, investing heavily in capacity before pivoting to profitability. Industry leaders have provided strong endorsements. GitHub's VP of Product noted excellence in complex code fixes across large codebases, while Cursor's CEO highlighted improvements in long-horizon tasks and difficult problems. Replit's President emphasized the extraordinary performance-to-cost ratio, and Cognition's CEO stated that Sonnet brings frontier-level reasoning in a more cost-effective package. The release includes several platform updates: adaptive and extended thinking capabilities, context compaction for longer conversations, and enhanced web search tools that automatically filter and process results. For Excel users, the add-in now supports MCP connectors, enabling integration with financial data providers like S&P Global and Bloomberg. Safety evaluations indicate Sonnet 4.6 maintains or improves upon previous models' safety standards, with researchers noting strong safety behaviors and no major alignment concerns. The model shows improved resistance to prompt injection attacks compared to its predecessor. This release positions Anthropic more competitively against coding-focused AI tools like GitHub Copilot, Cursor, and emerging agents like Devin. The combination of improved coding capabilities, computer use skills, and maintained pricing could accelerate adoption among developers and enterprises seeking cost-effective AI assistance for complex technical tasks.

Anthropic has unveiled Claude Sonnet 4.6, representing a substantial advancement in AI model capabilities that could significantly impact the competitive dynamics of coding assistants and computer automation tools. The release demonstrates meaningful progress across multiple domains while maintaining cost parity with previous versions.

The new model introduces comprehensive improvements in coding performance that have garnered strong positive feedback from early adopters. Testing within Claude Code revealed that users preferred Sonnet 4.6 over its predecessor approximately 70% of the time, with particular praise for enhanced context comprehension and more efficient code organization. Users reported that the model better consolidates shared logic rather than creating duplicate code, resulting in less frustrating extended coding sessions.

Perhaps more significantly, Sonnet 4.6 outperformed Claude Opus 4.5, Anthropic's previous flagship model from November 2025, in 59% of user comparisons. Early adopters noted reduced overengineering tendencies, decreased instances of false success claims, fewer hallucinations, and more consistent execution of multi-step tasks. This performance elevation brings capabilities that previously required premium Opus-class models to the more accessible Sonnet tier.

Computer use functionality has experienced dramatic improvements since Anthropic first introduced general-purpose computer interaction capabilities in October 2024. The OSWorld benchmark, which evaluates AI performance across real software applications including Chrome, LibreOffice, and VS Code, shows steady progress over sixteen months. Sonnet 4.6 now demonstrates human-level competency in complex tasks such as navigating intricate spreadsheets and completing multi-step web forms across multiple browser tabs.

The model's expanded 1-million token context window enables processing of entire codebases, comprehensive contracts, or extensive research collections in single requests. This capability proved particularly valuable in the Vending-Bench Arena evaluation, where Sonnet 4.6 developed sophisticated business strategies, initially investing heavily in capacity building before strategically pivoting to profitability optimization.

Industry feedback has been overwhelmingly positive across multiple sectors. GitHub's leadership highlighted the model's excellence in complex code fixes requiring extensive codebase searches. Cursor's executive team noted improvements in long-horizon tasks and challenging problem-solving scenarios. Replit's management emphasized the exceptional performance-to-cost ratio, while Cognition's leadership described Sonnet 4.6 as bringing frontier-level reasoning capabilities in a more economical package.

The platform introduces several technical enhancements including adaptive and extended thinking capabilities, context compaction for managing longer conversations, and improved web search functionality that automatically filters and processes results for better token efficiency. Excel integration has been enhanced with MCP connector support, enabling seamless integration with financial data providers such as S&P Global, LSEG, and PitchBook.

Safety considerations remain paramount, with comprehensive evaluations indicating that Sonnet 4.6 maintains or exceeds previous safety standards. Research teams noted strong safety behaviors and absence of major alignment concerns, while also documenting improved resistance to prompt injection attacks compared to earlier versions.

This release positions Anthropic more competitively against established coding tools and emerging AI agents, potentially accelerating enterprise adoption of AI-assisted development workflows. The combination of enhanced capabilities and maintained pricing structure could influence market dynamics across the AI coding assistant landscape.

Referenced links