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 to exact-match instead of substring-matching. Fixed voice dictation on macOS and auto-submit for languages without spaces (Japanese, Chinese, Thai). Fixed background agent daemons running unreachable when control socket fails to start.

Added autoMode.classifyAllShell setting to route all Bash/PowerShell commands through the auto-mode classifier, with denial reasons surfaced in transcripts and /permissions. Added claude_code.assistant_response OpenTelemetry log event for model responses. Added live file path autocomplete to bash mode (!). Fixed /model and client-data-gated UI showing stale state after /login.

Added /rewind support to resume conversations from before /clear was run. Fixed scroll position jumping during streaming responses. Fixed background agents resurrecting after being stopped. Fixed /voice showing generic "not available" message — now explains organization policy restrictions.

claude mcp login/logout — authenticate a configured MCP server from your shell instead of the interactive /mcp menu; clear stored credentials with claude mcp logout.

Shell mode responds to command output — run ! npm test and get an explanation of the failures without a second prompt. Set respondToBashCommands: false in settings.json to revert.

Background subagents surface permission prompts — instead of auto-denying, background agents now bubble up permission requests to the main session so you can approve them.

Streaming performance — streaming responses now use about 37% less CPU, and long-session memory growth from terminal output cache is reduced.

Sandbox credential blocking — new sandbox.credentials setting blocks sandboxed commands from reading credential files and secret environment variables.

Rate limits raised across the API. Claude Sonnet and Haiku rate limits now match Claude Opus at every usage tier. Usage tiers consolidated to three: Start, Build, and Scale — most organizations move to a higher tier, none receive lower limits. No action required; view current limits in the Claude Console.

Fast mode deprecated for Claude Opus 4.7, with removal on July 24, 2026. After removal, requests to claude-opus-4-7 with speed: "fast" will return an error. Migrate to fast mode for Claude Opus 4.8.

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

Claude Tag — Claude Code made multiplayer, async, and proactive across teams. Now in beta for Enterprise and Team plans. Writes 65% of Anthropic's product team's code, including most of Claude Tag itself. "Tag it in, it tags you back."

Computer Use in Claude Code — now in research preview on Pro/Max. Can open apps, click through UI, and test built work directly from the CLI without leaving your terminal.

Boris used Claude Code to decipher Linear A (a 3,500-year-old Cretan script) — results pending peer review. Also shipping Artifacts heavily for visual code explanations, system diagrams, animation previews, and dashboards. "Hasn't written a line of code himself in 2026."

Thariq on dynamic workflows — Claude can write its own harness on the fly, custom-built for each task. Also demoing control of Claude Code sessions via Telegram and Discord MCPs, messaging Claude from your phone. New /insights command summarizes projects and workflow suggestions from your past month of usage.