What's new in Claude

Added autoMode.classifyAllShell setting to route all Bash/PowerShell commands through the auto-mode classifier, plus live file path autocomplete in bash mode and automatic memory-pressure reaping for idle background shell commands.

Added /rewind support for resuming conversations from before /clear was run, and fixed multiple background agent and scroll positioning issues.

Bug fixes and reliability improvements across the board.

/cd command — move the current session to a new working directory mid-conversation without rebuilding the prompt cache.

Sub-agents can spawn sub-agents — background chains are now supported up to five levels deep, enabling deeply nested multi-agent orchestration.

--safe-mode — start Claude Code with all customizations (hooks, plugins, MCPs) disabled for clean troubleshooting sessions.

fallbackModel — configure up to three fallback models tried in order when your primary model is unavailable or rate-limited.

All major SDKs (Python, TypeScript, Go, Java, Ruby, PHP, C#) now support code_execution_20260120 — the code execution tool version with REPL state persistence and programmatic tool calling. Available on Claude Fable 5, Mythos 5, Opus 4.5+, and Sonnet 4.5+.

Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) have been retired — all requests now return an error. Migrate to Sonnet 4.6 and Opus 4.8 respectively.

Claude Fable 5 launched as the most capable widely released model, with a 1M token context window by default, 128k max output tokens, always-on adaptive thinking, and a stop_reason: "refusal" field. Claude Mythos 5 also launched for Project Glasswing participants.

Claude Tag debuted in Slack, bringing agentic capabilities from Claude Code into Slack channels. Rate limits on Claude Code were also doubled for all users. Earlier this month, a structural credits change decoupled human-interactive usage from programmatic/agentic usage starting June 15.

Claude Fable 5 launched June 9 as the new flagship model. Claude Design token limits were doubled across Pro, Max, Team, and Enterprise plans. Claude Code rate limits doubled.

Boris reflected on Claude Code's deep customizability — hooks, plugins, LSPs, MCPs, skills, custom agents, effort levels — as a key driver of why engineers love it. He noted he hasn't written a line of code himself in 2026, running 5 parallel Claude Code instances across git worktrees.

Thariq announced dynamic workflows — Claude now writes its own multi-agent harness on the fly, custom-built per task. Called it "the biggest upgrade to Claude Code's capabilities since skills and subagents." Also shared a spec-based dev technique: have Claude interview you with AskUserQuestionTool, then execute the spec in a fresh session.