CHAPTERS
- 0:00 – 0:30
Claude Sonnet 4.5 release: top coding performance and long-horizon focus
The update opens with Claude Sonnet 4.5 now broadly available and positioned as a leading coding model. It highlights benchmark leadership and the ability to stay focused on complex tasks for extended periods.
- •Sonnet 4.5 is available wherever you use Claude
- •Leads SWE-bench Verified with a 77.2% score
- •Demonstrated focus on complex tasks for 30+ hours
- •Framed as a major win for developer workflows
- 0:30 – 1:00
Broader capability gains + Claude for Chrome availability
Beyond coding, the speaker points to improvements in reasoning, math, and computer-use skills. They cite OSWorld gains and direct viewers to try the expanded Claude for Chrome extension.
- •Substantial gains in reasoning, math, and computer use
- •OSWorld improvement from 42% to 61% in four months
- •Claude for Chrome extension expanded to everyone on the waitlist
- •Call to action: try it at claude.ai/chrome
- 1:00 – 1:31
Claude Code in the IDE: native VS Code extension (beta)
Claude Code gets a native VS Code extension that brings the experience directly into developers’ preferred IDE workflows. The extension emphasizes visibility into changes via a sidebar and inline diffs.
- •Native VS Code extension integrates Claude Code into the IDE
- •Designed for developers who prefer IDEs over terminals
- •Dedicated sidebar shows changes in real time
- •Inline diffs for edits made by Claude
- •Available in beta on the VS Code Marketplace
- 1:31 – 2:01
Claude Code 2.0 terminal refresh: UI, visibility, and prompt history
Terminal users aren’t left behind: Claude Code 2.0 introduces a refreshed terminal UI. The update focuses on improved interface clarity and better navigation of past prompts.
- •Claude Code updated to version 2.0
- •Refreshed terminal interface features
- •Improved status visibility
- •Searchable prompt history
- 2:01 – 2:23
Checkpoints and /rewind: safe rollbacks for large Claude-driven edits
A new checkpoints feature enables quick rollback to prior states during big tasks. Users can restore code, conversation, or both, with clear boundaries on what checkpoints cover.
- •New checkpoints feature for confident large-task execution
- •Use /rewind or double-press Escape to activate
- •Restore code, conversation, or both
- •Checkpoints apply only to Claude edits (not user edits or bash commands)
- •Recommended to use alongside version control
- 2:23 – 2:31
Faster control over “thinking” + real-time usage tracking
Claude Code adds quality-of-life improvements: toggling thinking is simplified and preferences persist. Users can also monitor usage in real time from the CLI or within the Claude app.
- •Enable/disable thinking with the Tab key
- •Thinking preference is saved across sessions
- •Track usage in real time with /usage
- •In the Claude app: Settings → Usage
- 2:31 – 3:02
Claude API: context editing to extend agent runtime under token limits
On the API side, new context editing helps agents operate longer without manual cleanup. It automatically removes stale tool calls/results when nearing token limits while preserving conversational continuity.
- •New API capability: context editing
- •Automatically clears stale tool calls and tool results near token limits
- •Preserves conversation flow while trimming context
- •Helps agents run longer without manual user intervention
- 3:02 – 3:17
Claude API: memory tool for persistent, client-side agent state
The memory tool lets Claude store and retrieve information outside the context window using a file-based system. This supports persistence across conversations while keeping storage client-side in your infrastructure.
- •New API capability: memory tool
- •File-based memory directory stored in your infrastructure (client-side)
- •Claude can create/read/update/delete memory files
- •Persists across conversations beyond the context window
- •Analogy: like a claude.md for your agent API
- 3:17 – 3:32
Examples and resources: Catan demo and cookbooks
The speaker points to concrete examples and learning materials that show how to use context editing and memory effectively. Viewers are directed to a demo video and supporting cookbooks.
- •Example referenced: “Claude Plays Catan” video
- •Cookbooks provided to demonstrate new capabilities
- •Focus on practical guidance for leveraging context editing and memory
- 3:32 – 4:02
Claude Agent SDK (renamed): the tools behind Claude Code for your own agents
The Claude Code SDK is renamed to the Claude Agent SDK and packaged as a way to build custom agents using the same underlying systems. It exposes core tools, context management, and permission frameworks used in Claude Code.
- •Renamed from Claude Code SDK to Claude Agent SDK
- •Provides the core tools that power Claude Code
- •Includes context management systems
- •Includes permissions frameworks
- •Informed by learnings from the last six months
- 4:02 – 4:33
Claude app file generation: Excel, PowerPoint, Word, and PDF outputs
In the Claude app, Claude can now analyze data, visualize insights, and generate downloadable files in common business formats. The feature is available as a preview for paid plans.
- •Claude can use code to analyze data and create files
- •Generate Excel spreadsheets via natural language prompts
- •Create PowerPoint presentations, draft Word docs, or export PDFs
- •Files are downloadable for immediate use
- •Available in preview for all paid plans
