Claude Code v2.0.0

What: New default model claude-sonnet-4-5-20250929 with improved performance for daily coding tasks

How to use:

# Sonnet 4.5 is now the default - no changes needed
claude

# Or explicitly specify it
claude --model sonnet45

Details:

• Available across all platforms (first-party API, Bedrock, Vertex AI)
• Automatic migration from previous default models
• Described as "Smartest model for daily use"
• Supports 1M token context window variant: claude-sonnet-4-5[1m]
• Evidence: U10 at line 367516, GM1() at line 368315, model detection logic at line 340617 in v2.0.0

What: View your subscription plan usage limits and current consumption

How to use:

/usage

Details:

• Shows current 5-hour session usage
• Displays 7-day usage across all models
• Shows 7-day Opus-specific usage
• Visual progress bars with percentage indicators
• Reset time countdown
• Only available for subscription plans (API key users see informational message)
• Evidence: OXB command definition at line 416198, usage API endpoint at line 402408, UI components E3B() and WS6() at lines 402479-402532 in v2.0.0

What: Complete overhaul with adaptive layout, branded "Clawd" mascot, and dynamic content feeds

Features:

• Clawd mascot: Custom ASCII art character with theme-aware rendering
• Three responsive layouts: Adapts to terminal width (compact/horizontal/vertical)
• Dynamic feeds: Recent activity with timestamps, release notes, getting started tips
• Personalized greeting: Shows user's display name when available
• Model & billing info: Quick view of current settings

Details:

• Automatically adjusts to terminal size and capabilities
• Special rendering optimizations for Apple Terminal
• Supports both light and dark themes
• Shows recent conversation history
• Displays latest changelog entries
• Contextual tips based on project state
• Evidence: Main welcome screen iOB() at line 429317, Clawd mascot PK5() at line 429231, feed system gOB() at line 429168, splash screen Pc1() at line 441300 in v2.0.0

What: Simplified thinking trigger system with new programmatic control

How to use:

# Use "ultrathink" anywhere in your prompt for maximum thinking
claude
> Please ultrathink about this architectural design

# Toggle thinking with tab key during interactive sessions
# Or use /t command

Details:

• Streamlined from multilingual pattern matching to single "ultrathink" keyword
• Binary mode: ULTRATHINK (31,999 tokens) or NONE
• New thinkingMetadata API for programmatic control without text triggers
• Removed complex multi-level system (HIGHEST/MIDDLE/BASIC) in favor of simpler on/off
• Thinking state notifications show "Thinking on/off (tab to toggle)"
• Evidence: Pattern at line 363628, functions Y61() at line 363667, Y09() at line 363644, le1() at line 363671, metadata handling at line 363649 in v2.0.0

What: Tabbed interface combining status, configuration, and usage in one place

How to use:

/status  # Opens with Status tab
/config  # Opens with Config tab  
/usage   # Opens with Usage tab

Details:

• Tab navigation to cycle between Status/Config/Usage panels
• Status tab shows system diagnostics, IDE connection, MCP servers, memory usage
• Config tab provides interactive settings management
• Usage tab displays subscription limits (new in 2.0.0)
• Unified interface reduces command clutter
• Evidence: Tabbed component L01() at line 394930, status dialog h01() at line 402533 in v2.0.0

What: Enhanced session history deduplication and conflict detection

Details:

• Prevents duplicate entries when fetching session history
• Tracks already-fetched UUIDs per session to avoid redundant storage
• Improved conflict resolution with UUID mismatch detection
• Better error messages for concurrent modification scenarios
• Evidence: Deduplication logic in F8B() at line 397200, UUID tracking with Cv1 map in v2.0.0

What: Shows currently active file or selected text from connected IDE

Display:

• "⧉ In filename.ts" when a file is active
• "⧉ 15 lines selected" when text is selected

Details:

• Only appears when IDE is connected and has an active context
• Helps confirm Claude knows which file you're working in
• Updates automatically as you switch files
• Evidence: aLB() function at line 423498 in v2.0.0

What: Improved notification queue management with invalidation support

Details:

• Notifications can now invalidate previous notifications by key
• Better handling of immediate-priority notifications
• Prevents notification spam by filtering duplicates
• Queue management ensures only relevant notifications are shown
• Evidence: Enhanced xG() notification hook at line 363012 with invalidates support at lines 363225-363244 in v2.0.0

What: Context window display now shows reserved space for auto-compact and output tokens

Details:

• New "Reserved (autocompact + output tokens)" category in /context display
• Shows available tokens after accounting for auto-compact threshold
• Helps users understand actual usable context space
• More accurate free space calculations for subscription plans
• Evidence: L3B and CS6 variables at lines 402598 and 403199, calculation logic in y3B() at line 403320 in v2.0.0

What: Visual feedback when Claude uses extended thinking

Display:

∴ Thinking…  (while active)
∴ Thought for 12s (ctrl+o to show thinking)  (after completion)

Details:

• Real-time indicator during thinking phase
• Shows total thinking duration after completion
• Reminds users how to view thinking content
• Only appears when thinking is actually used
• Evidence: m3B() function at line 404483 in v2.0.0

What: Improved file path display in narrow terminals

Details:

• Smart truncation preserves first and last path components
• Uses middle ellipsis (…) to show structure
• Handles root paths and edge cases gracefully
• Example: /very/long/path/to/file.ts → /very/…/to/file.ts
• Evidence: MI1() function at line 405562 in v2.0.0

What: Clearer model naming in UI

Changes:

• "Sonnet 4.5" for the new default model
• "Sonnet 4.5 (with 1M token context)" for extended context variant
• Consistent capitalization and formatting
• Evidence: UeA() function at line 368322, JM() at line 368327, lJA() at line 367531 in v2.0.0

What: Better file information when restoring changes

Details:

• Shows specific filenames instead of generic "files changed"
• Lists up to 2 files by name, then "and N other files"
• More informative change summaries
• Evidence: uf6() function at line 414892 in v2.0.0

What: Welcome screen shows recent conversations with timestamps

Details:

• Displays last 3 non-sidechain conversations
• Filters out apology-heavy sessions
• Shows conversation summary or first prompt
• Relative timestamps (e.g., "2h ago", "yesterday")
• Quick context for resuming work
• Evidence: _GB() function at line 405621, iR0() at line 429186 in v2.0.0

What: Automatic model upgrade for existing users

Details:

• Users with explicit model settings are migrated to Sonnet 4.5
• Migration timestamp tracked to show "Model updated to Sonnet 4.5" notification
• One-time migration prevents repeated notifications
• Graceful handling of users who prefer other models
• Evidence: v4Q() migration function at line 443290, notification in NRB() at line 430680 in v2.0.0

What: Prompts to activate Claude subscription if available

Details:

• Shows up to 3 times for eligible users
• "Use your existing Claude [plan] with Claude Code · /login to activate"
• Tracks notification count to avoid annoyance
• Only shown to users with available subscriptions
• Evidence: MRB() notification at line 430699, subscription check sK5() referenced in v2.0.0

What: Streamlined from multilingual multi-level system to single trigger

Previous (v1.0.128):

• 8 languages supported (English, Japanese, Chinese, Spanish, French, German, Korean, Italian)
• 4 thinking levels: HIGHEST (31,999 tokens), MIDDLE (10,000), BASIC (4,000), NONE (0)
• Dozens of trigger phrases per language
• Complex pattern matching with negation filtering

Now (v2.0.0):

• English-only "ultrathink" trigger
• 2 levels: ULTRATHINK (31,999 tokens) or NONE (0)
• Simple case-insensitive regex matching
• New thinkingMetadata API for programmatic control

Implications:

• More predictable behavior
• Easier to remember single trigger word
• Programmatic control via metadata for advanced use cases
• Evidence: Removed v19 patterns at line 363665 in v1.0.128, new simplified system with NU1 at line 363628 and functions at lines 363644-363671 in v2.0.0

What: Removed MultiEdit from available tools array

Details:

• MultiEdit tool completely removed from codebase
• Tool arrays updated from ["Edit", "MultiEdit", "Write", "NotebookEdit"] to ["Edit", "Write", "NotebookEdit"]
• Edit tool handles all file editing use cases
• Reduces complexity and potential confusion between Edit and MultiEdit
• Evidence: Tool array HK5 at line 428943 in v1.0.128 vs IK5 at line 428118 in v2.0.0

What: Simplified state update logic removes redundant null checks

Details:

• Model selection updates no longer check if oldState is null before comparison
• Assumes initial state is always available
• Cleaner conditional logic
• Added thinking state persistence to user settings
• Evidence: Function Bp() at line 443042 in v1.0.128 vs df() at line 443128 in v2.0.0

What: Reduced thinking notification color levels

Details:

• Removed medium and low thinking level colors
• Now only "none" (text) and "high" (claude) colors
• Matches simplified binary thinking system
• Evidence: Variables cT and aa at lines 363617-363618 in v1.0.128 vs LU1 and $WA at lines 363586-363587 in v2.0.0

What: The MultiEdit tool has been removed entirely

Reason: The Edit tool provides all necessary file editing capabilities, and having both tools caused confusion.

Migration: Use the Edit tool for all file modifications. For multiple edits to the same file, Claude will use Edit multiple times as needed.

Evidence: Variable sH = "MultiEdit" at line 365916, tool registration in arrays at line 428943, prompts mentioning MultiEdit at lines 440925-440941 in v1.0.128 - all removed in v2.0.0

What: Support for non-English thinking trigger phrases removed

Affected Languages:

• Japanese (熟考, 深思, etc.)
• Chinese (深思, 仔细思考, etc.)
• Spanish (piensa más, etc.)
• French (réfléchis, etc.)
• German (denk nach, etc.)
• Korean (심사숙고, 깊이 생각, etc.)
• Italian (pensa profondamente, etc.)

Replacement: Use "ultrathink" in English for maximum thinking tokens

Evidence: Variable v19 with multilingual patterns at line 363665 in v1.0.128 - removed in v2.0.0, replaced with single English pattern G09 at line 363628

What: Old ASCII banner and static welcome screen removed

Removed Components:

• Large ASCII art "CLAUDE CODE" logo (lGB() at line 406445)
• Simple header with star icon (DN0() at line 406867)
• Static environment overrides display (_A1() at line 407407)

Replacement: New adaptive welcome screen with Clawd mascot and dynamic feeds

Evidence: Functions lGB(), DN0(), _A1() in v1.0.128 completely replaced by new system in v2.0.0

What: Removed "Opus Plan Mode" model fallback strategy

Details:

• Previously: "Use Opus 4.1 in plan mode, Sonnet 4 otherwise"
• This specific fallback configuration is no longer available
• Standard model selection and rate limit fallback still work
• Evidence: Variable D31 at line 368627, function U31() at line 368577 in v1.0.128 - removed in v2.0.0

What: Removed several project-level config management functions

Removed:

• Ym1() - Add items to array config (line 439101)
• YPB() - Remove items from array config (line 439140)
• VPB() - Set config value (line 439430)
• KPB() - Delete config key (line 439488)
• FPB() - Get config value (line 439413)
• zPB() - List config (line 439507)

Reason: Config management migrated to settings.json system

Migration: Use .claude/settings.json for project configuration instead of CLI commands

Evidence: Functions at lines 439101-439507 in v1.0.128 marked as removed in diff

What: Removed dedicated function for creating post-tool feedback system messages

Details:

• Function kv1(A, B) removed - created system messages for tool hook feedback
• Feedback mechanism still exists but implementation changed
• Evidence: Function kv1() at line 397255 in v1.0.128 vs different implementation in v2.0.0

What: Fixed duplicate session log entries when fetching history

Issue: When resuming sessions, the same entries could be persisted multiple times

Fix: Added UUID tracking per session to skip already-fetched entries

Impact: Cleaner session history, reduced storage overhead, faster session loading

Evidence: New deduplication check at start of F8B() function at line 397200 in v2.0.0

What: Improved responsive layout calculations for welcome screen

Details:

• Better handling of very narrow terminals (< 70 columns)
• Fixed edge cases in path truncation
• More accurate column calculations for multi-panel layouts
• Evidence: Layout functions SGB() at line 405537, yGB() at line 405541 in v2.0.0

1. MultiEdit tool removed - Extensions or scripts using MultiEdit must switch to Edit tool
2. Non-English thinking triggers removed - Only "ultrathink" works now
3. Project config CLI commands removed - Use settings.json instead
4. Opusplan mode removed - Adjust model selection strategy if using this

• MultiEdit → Edit: Simply use Edit tool; Claude will apply multiple edits as needed
• Multilingual thinking → ultrathink: Replace language-specific triggers with "ultrathink"
• Config commands → settings.json: Move allowedTools, ignorePatterns to .claude/settings.json permissions object
• Model selection: Sonnet 4.5 is now default; explicit model selections are preserved

• Version: 2.0.0
• Previous Version: 1.0.128
• Release Date: 2025 (based on model identifiers)
• Major Version Bump Reason: Breaking changes (removed tools, removed CLI commands, simplified thinking system)