Amid the intensifying competition in AI-assisted development tools, Anthropic officially released a major version update (v2.0) of Claude Code on September 29, along with the launch of the Claude Sonnet4.5 model.
Claude Code Enhancements Reshape the AI Programming Ecosystem
This update focuses on autonomy and integration, aiming to transform AI from a mere code assistant into a reliable "programming partner." According to the official announcement, Claude Code now supports handling longer and more complex tasks, and it seamlessly collaborates within terminal and IDE environments. The update quickly sparked discussions in the developer community, seen as a direct response to competitors like OpenAI Codex.
Sonnet4.5 Model: A Masterpiece in Coding and Agent Construction
Claude Sonnet4.5 is positioned as the best coding model globally, scoring 61.4% on the OSWorld benchmark, an increase of nearly 20 percentage points over its predecessor, Sonnet4. It performs well in real computer tasks, reasoning, and math problems, especially in building complex agent systems. It can autonomously run continuous coding tasks for over 30 hours without being interrupted by context loss.
In addition, Sonnet4.5 enhances its tool coordination capabilities, including code execution, file creation (such as spreadsheets, slides, and documents), and visualization support. These features are directly integrated into the Claude chat interface, allowing users to complete data analysis or prototype building without switching tools. The pricing remains unchanged, at $3 per million input tokens and $15 per million output tokens, ensuring high performance without additional costs.
Claude Code v2.0: Dual Innovations in Terminal and IDE
The core of this update lies in the interface and feature improvements of Claude Code. First, the terminal interface (CLI) has undergone a complete redesign, featuring a smoother UI that supports parallel processes and real-time feedback. Developer feedback indicates that its response speed has improved by about three times, reducing previous lag issues.
Second, the native VS Code extension has entered beta testing. This plugin integrates Claude Code into the IDE sidebar, offering inline diffs and graphical interactions. Users can view AI-modified code changes in real time and revert with one click, significantly improving collaboration efficiency. This extension is currently only available for VS Code, while third-party IDEs like Cursor will need to wait for future compatibility.
Checkpoint Mechanism: A "Safety Net" for Autonomous Operations
Claude Code introduces a revolutionary checkpointing feature, allowing the AI to automatically save its state during long tasks. Users can roll back to the last checkpoint at any time, restoring code or conversation without affecting manual editing or command-line operations. This feature, combined with version control tools (like Git), is suitable for large-scale project development, preventing the risk of AI "running out of control."
Additionally, the "Tab to Think" shortcut optimizes the thinking mode, allowing users to trigger the AI's deep reasoning process simply by pressing Tab. The new mascot, "Clawd" (a cute crab), adds a fun touch to the interface, enhancing user engagement.
API and Ecosystem Expansion: Upgrading the Developer Toolbox
To support enterprise applications, Anthropic simultaneously updated the Claude API, adding context editing (automatically cleaning up outdated information) and memory tools (storing data beyond the window limit). The Claude Agent SDK (formerly the Claude Code SDK) is now open to team users, providing core tools and permission frameworks to customize the agent experience.
GitHub Copilot has included Sonnet4.5 in its public test, supporting Pro and higher plan users to use it in chat, edit, and agent modes. The browser extension Claude for Chrome has also been extended to Max subscribers, further integrating AI capabilities into daily workflows.
Safety and Compliance: ASL-3 Level Protection
All updates were developed under the AI Safety Level 3 (ASL-3) framework, strengthening defenses against prompt injection and reducing hallucinations and compliance behaviors. Enterprise users can integrate data retention and issue marking functions via the Compliance API to ensure compliance.
Looking Ahead: Accelerating the New Era of AI Programming
This iteration of Claude Code marks a leap of AI from an auxiliary tool to an autonomous agent. Developers can immediately experience these features on Claude.ai or via the API, with paid plan users getting priority access. Anthropic stated that future efforts will focus on longer autonomous tasks and multimodal integration. AIbase will continue to track developments in this field; stay tuned.
