What's new in Claude

Added CLAUDE_CODE_DISABLE_MOUSE_CLICKS env var to disable mouse interactions in fullscreen mode while keeping wheel scroll. Fixed hook matchers with hyphenated identifiers (e.g. code-reviewer) accidentally substring-matching — they now exact-match. Fixed voice dictation on macOS capturing silence after the default input device changes, and fixed auto-submit never firing for languages written without spaces (Japanese, Chinese, Thai).

Added autoMode.classifyAllShell setting to route all Bash/PowerShell commands through the auto-mode classifier, with denial reasons now surfaced in transcripts and /permissions. Added live file path autocomplete in bash mode (!). Added automatic memory-pressure reaping for idle background shell commands.

Added /rewind support to resume a conversation from before /clear was run. Fixed scroll position jumping to the bottom while reading earlier output during a streaming response. Fixed background agents resurrecting after being stopped.

claude mcp login / logout — authenticate a configured MCP server directly from your shell without going through the interactive /mcp menu; clear stored credentials later with claude mcp logout <name>.

Shell mode responds to command output — running ! npm test now triggers Claude to explain the output automatically without a second prompt. Set respondToBashCommands: false in settings.json to revert to the previous behavior.

/rewind after /clear — resume a conversation from before you cleared it; Claude reloads the prior context window so nothing is truly lost.

Background subagents surface permission prompts — instead of auto-denying blocked actions, background agents now bubble the prompt up to the main session so you can approve in real time.

Fast mode removed for Claude Opus 4.6. Requests with speed: "fast" to claude-opus-4-6 no longer run at fast speed or premium pricing — they now run at standard speed and are billed at standard rates. Migrate to fast mode on Claude Opus 4.8 to continue using fast mode.

API rate limits raised. Claude Sonnet and Haiku rate limits now match Claude Opus at every usage tier. Usage tiers consolidated into three: Start, Build, and Scale — most organizations move up, none move down. No action required; check current limits in the Claude Console.

Code execution tool across all major SDKs. Python, TypeScript, Go, Java, Ruby, PHP, and C# now support code_execution_20260120 with REPL state persistence. No beta header required. Available on Fable 5, Mythos 5, Opus 4.5+, Sonnet 4.5+.

Claude Tag is in beta for Enterprise and Team plans — Claude Code made multiplayer, async, and proactive. Anthropic's internal product team is shipping 65% of their code using it, including the code that built Claude Tag itself.

Sharing workflow tips: running 3–5 git worktrees simultaneously with separate Claude Code instances in parallel, shipping 20–30 PRs a day. Philosophy: update CLAUDE.md after every correction so Claude doesn't repeat mistakes. Also: PRDs are dead on the Claude Code team — prototypes replaced them.

Highlighted Dynamic Workflows (Claude writes its own orchestration harness on the fly), Claude Code Channels (control sessions via Telegram or Discord, message from your phone), and the new /btw command for side-chain conversations while Claude is working.