Claude Coded: Sonnet 4.5, Claude Code 2.0, and more.

Key Concepts

• Claude Sonnet 4.5
• SweetBench
• OS World
• Claude for Chrome extension
• Claude Code (VS Code extension, Terminal UI, Checkpoints, Thinking, Usage tracking)
• Context Editing (Claude API)
• Memory Tool (Claude API)
• Claude Agent SDK
• Claude App (Data analysis, file creation, visualization)
• IDE (Integrated Development Environment)
• Inline Diffs
• Token Limits
• Clientside Storage
• CRUD Operations

---

Claude Sonnet 4.5: The Best Coding Model

The update introduces Claude Sonnet 4.5, now available across all Claude platforms, asserting its position as the "best coding model in the world." This claim is substantiated by its performance on SweetBench, where it leads with a verified score of 77.2%. Beyond benchmarks, Sonnet 4.5 has demonstrated remarkable endurance, staying focused on complex coding tasks for "well over 30 hours straight." The improvements extend beyond just code, showing substantial gains in reasoning, math, and general computer use. On OS World, a test evaluating an AI's ability to use a computer like a human, Claude's score significantly jumped from 42% four months ago to over 61% currently.

Enhanced User Interaction: Claude for Chrome Extension

The practical application of Claude's improved computer usage capabilities is showcased through the recently launched Claude for Chrome extension. This extension has been expanded to include "everybody that was on the wait list" and is available at cloud.ai/chrome, allowing users to experience Claude's advanced interaction firsthand.

Claude Code: Developer Tooling Advancements

Claude Code has received a suite of new improvements designed to streamline developer workflows:

• Native VS Code Extension: A new native VS Code extension integrates Claude Code directly into the IDE, catering to developers who prefer programming in an IDE over the terminal. This extension features a dedicated sidebar panel that displays Claude's changes in real-time through inline diffs, providing immediate visual feedback. It is currently in beta and accessible via the VS Code marketplace.
• Refreshed Terminal UI (v2.0): For terminal users, Claude Code version 2.0 brings an updated interface with improved status visibility and a searchable prompt history.
• New Checkpoints Feature: A significant addition is the Checkpoints feature, enabling developers to confidently run large tasks with the ability to instantly roll back to a previous state if needed. This feature is activated using the /re command or by double-hitting the escape key, allowing users to restore code, conversation, or both to a prior state. It's important to note that checkpoints only apply to edits made by Claude, not user edits or bash commands, thus recommending its use in conjunction with a version control system.
• Thinking Toggle: The "thinking" process can now be enabled or disabled with a single tap key, and this preference is saved across sessions.
• Real-time Usage Tracking: Users can track their Claude Code usage in real-time using the /usage command in the terminal or by navigating to Settings > Usage within the Claude app.

Claude API: Enabling Greater Agent Complexity

Two new capabilities have been introduced to the Claude API, designed to empower agents to handle even greater complexity:

• Context Editing: This feature automatically clears "stale tool calls and results" from within the context window when approaching token limits. As an agent executes tasks and accumulates tool results, context editing intelligently removes outdated content while preserving the conversation flow, effectively extending how long an agent can run without manual user intervention.
• Memory Tool: The Memory Tool allows Claude to store and consult information outside of the context window through a file-based system. Claude can perform CRUD operations (create, read, update, delete) on files within a dedicated memory directory. This directory is stored in the user's infrastructure, is "entirely clientside," and persists across conversations, akin to having a claw.md file for an agent API. Examples of these capabilities are demonstrated in the "Claude plays Katan" video, and specific cookbooks are available to guide users on leveraging these new features.

Claude Agent SDK: Building Custom Agents

The Claude Code SDK has been renamed to the Claude Agent SDK. This SDK provides developers with access to the same core tools, context management systems, and permissions frameworks that power Claude Code, facilitating the creation of custom agents. It incorporates extensive learnings gathered over the past six months to offer a robust framework for agent development.

Claude App: Data Analysis and File Generation

The Claude app has been significantly enhanced, allowing Claude to use code for advanced data analysis, file creation, and insight visualization. Users can now prompt Claude using natural language to generate familiar file formats such as Excel spreadsheets, PowerPoint presentations, Word documents, or PDF files, which can be instantly downloaded and used. This capability is currently available in preview for all paid plans.

---

Conclusion

This comprehensive update across the Claude ecosystem underscores a significant leap in AI capabilities, particularly in coding, reasoning, and computer interaction. Key takeaways include the superior performance of Claude Sonnet 4.5, substantial improvements in developer tooling with Claude Code (especially the VS Code extension and Checkpoints feature), and powerful new API features like Context Editing and the Memory Tool that enable the development of more complex and persistent AI agents. Furthermore, the Claude App now offers advanced data analysis and file generation, making Claude a more versatile and actionable tool for a broader range of users. The focus remains on enhancing efficiency, handling complexity, and providing intuitive user experiences for both developers and end-users.