Skip to content
AnthropicAnthropic

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

We're covering everything new with Claude for developers, including the launch of Claude Sonnet 4.5, major updates to Claude Code, powerful new API capabilities, and exciting features in the Claude app. Join the Claude Developer Discord - https://anthropic.com/discord Check out the Claude Code Docs - https://docs.claude.com/en/docs/claude-code/overview Check out the Memory Cookbook - https://github.com/anthropics/claude-cookbooks/blob/main/tool_use/memory_cookbook.ipynb 00:06 Claude Sonnet 4.5 - Leading on SWE-bench and can focus for 30+ hours on complex tasks 01:03 Claude Code VS Code Extension - Now available in beta on the VS Code Marketplace 01:30 Claude Code 2.0 - Refreshed terminal UI 01:45 Claude Code Checkpoints - Roll back to previous states with /rewind 02:23 Usage Tracking - Monitor your usage in real-time with /usage command 02:35 Claude API - Context editing and memory tools for building sophisticated agents 03:37 Claude Agent SDK - Build your own agents with the tools that power Claude Code 03:59 Claude File Creation - Generate Excel, PowerPoint, Word, and PDF files in the Claude app

Oct 6, 20254mWatch on YouTube ↗

CHAPTERS

  1. Claude Sonnet 4.5 released: top coding performance and long-task focus

    The update opens with Claude Sonnet 4.5 now available across Claude surfaces, positioned as Anthropic’s best coding model. It highlights benchmark leadership and the model’s ability to stay focused on complex work for extended periods.

    • Claude Sonnet 4.5 is broadly available “wherever you get your Claude”
    • Leads SWE-bench Verified with a 77.2% score
    • Demonstrated ability to stay focused on complex tasks for 30+ hours
    • Positioned as a major upgrade for developers and coding workflows
  2. Beyond code: gains in reasoning, math, and computer-use (OSWorld)

    The video emphasizes that improvements aren’t limited to programming—core reasoning and tool/computer interaction also advanced. OSWorld results are used to quantify progress in “using a computer like a human.”

    • Substantial gains noted in reasoning, math, and computer use
    • OSWorld score increased from 42% to 61% in about four months
    • Frames OSWorld as a practical measure of real computer-operation ability
    • Sets context for using Claude in more agentic, multi-step tasks
  3. Claude for Chrome extension expanded to waitlist users

    Anthropic points viewers to a hands-on way to see computer-use improvements through the Claude for Chrome extension. Access has been expanded to everyone who was previously on the waitlist.

    • Claude for Chrome recently launched and now expanded to waitlist users
    • Encourages users to try it to observe computer-use capabilities firsthand
    • Provides URL: claude.ai/chrome
    • Connects improved OSWorld performance to a real product experience
  4. Claude Code in the IDE: native VS Code extension (beta)

    A major Claude Code update is a native VS Code extension that brings Claude directly into the IDE. It targets developers who prefer IDE workflows over terminal-based tooling and adds real-time visibility into edits.

    • VS Code extension integrates Claude Code directly into the IDE
    • Dedicated sidebar shows Claude’s changes with inline diffs
    • Designed for developers who prefer IDEs over terminals
    • Available in beta via the VS Code Marketplace
  5. Claude Code 2.0: refreshed terminal UI and better visibility

    Terminal users also get improvements with a refreshed UI as Claude Code bumps to version 2.0. The update focuses on interface polish, clearer status indicators, and better retrieval of past prompts.

    • Claude Code updated to version 2.0
    • Refreshed terminal interface features
    • Improved status visibility for what Claude is doing
    • Searchable prompt history added
  6. Checkpoints and safe rollback with /rewind (plus limitations)

    A new checkpoints feature enables quick rollback to earlier states during large tasks, reducing risk when letting Claude make broad changes. The video also clarifies important constraints and recommends pairing with version control.

    • Checkpoints allow instant rollback to a previous state
    • Activate via /rewind or double-press Escape
    • Restore code, conversation, or both from a prior checkpoint
    • Only applies to edits made by Claude (not user edits or bash commands)
    • Recommended to use alongside a version control system
  7. Faster workflow controls: toggle “thinking” and keep preferences

    Claude Code’s “thinking” mode becomes easier to manage with a single-key toggle. The notable quality-of-life improvement is that the preference persists across sessions.

    • “Thinking” can be enabled/disabled with the Tab key
    • Preference is saved across sessions
    • Aims to streamline iterative coding sessions
    • Reduces repeated configuration during daily use
  8. Usage tracking in real time: /usage and app settings

    The update introduces clearer visibility into consumption and limits by adding a real-time usage command. Users can also find usage data within the Claude app settings.

    • Use /usage in Claude Code to monitor usage in real time
    • Usage view also available in the Claude app under Settings → Usage
    • Helps manage capacity during longer agentic or coding sessions
    • Improves transparency for developers running frequent tasks
  9. Claude API: context editing to extend long-running agents

    On the API side, context editing helps agents operate longer by removing stale tool calls/results as token limits approach. This is framed as a way to preserve conversation flow while reducing manual intervention.

    • Context editing automatically clears stale tool calls/results when nearing token limits
    • Preserves conversation continuity while trimming outdated content
    • Extends how long an agent can run without manual user cleanup
    • Targets complex, multi-step agent workflows
  10. Claude API: memory tool for persistent, client-side agent storage

    A new memory tool allows Claude to store and consult information outside the context window using a file-based system in the client’s infrastructure. This enables persistence across conversations while remaining entirely client-side.

    • Memory tool stores information outside the context window via files
    • Claude can create/read/update/delete files in a dedicated memory directory
    • Storage is in your infrastructure, client-side, and persists across conversations
    • Compared to having a persistent “claude.md” for an agent
  11. Claude Agent SDK (renamed): build agents with Claude Code’s core tools

    The SDK formerly known as the Claude Code SDK is renamed to the Claude Agent SDK. It exposes the tooling and frameworks that power Claude Code so developers can build their own agents with similar capabilities.

    • Claude Agent SDK renamed from Claude Code SDK
    • Provides access to core tools, context management, and permissions frameworks
    • Based on learnings from the last six months of development
    • Positioned as the foundation for building custom agents
  12. Claude app: generate downloadable Office/PDF files via natural language

    The video closes with a Claude app capability: using code to analyze data, visualize insights, and create common document formats. Users can request files in natural language and download them immediately, now available in preview for paid plans.

    • Claude can analyze data, create files, and visualize insights in familiar formats
    • Supports Excel, PowerPoint, Word, and PDF generation
    • Prompted via natural language and produces instantly downloadable files
    • Available in preview to all paid plans

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.