What's new in Claude

Stream-stall hint updated to "Waiting for API response · will retry in …" (was "No response from API · Retrying in …") and now triggers after 20 seconds of silence instead of 10.

Auto mode safety tightened: destructive git commands are now blocked, deprecation warnings added for retired models, and multiple issues with thinking blocks and WebSearch resolved.

New /config key=value syntax for settings, sandbox.allowAppleEvents on macOS, and long paragraphs now stream line-by-line instead of all at once.

/cd command: Move the current session to a new working directory mid-conversation without rebuilding the prompt cache — useful when jumping between repos in a long session.

Recursive sub-agents: Sub-agents can now spawn their own sub-agents; background chains are capped at five levels deep to prevent runaway trees.

--safe-mode: Starts Claude Code with all customizations disabled — a clean-slate mode for troubleshooting broken hooks, plugins, or settings.

fallbackModel: Configures up to three fallback models tried in order if the primary model is unavailable or rate-limited.

Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) are retired — all requests now return an error. Upgrade to Sonnet 4.6 or Opus 4.8.

Code execution tool now supports code_execution_20260521, which surfaces the 90-second per-cell limit in the tool description so Claude can budget long-running cells.

Web search and web fetch tools add a response_inclusion parameter to drop consumed result blocks from the API response, trimming token costs in agentic workflows. No beta header required.

Doubled rate limits on Claude Code are now live, alongside raised API limits for Claude Opus. The 5-hour and weekly counters were also reset for all paid plans.

Claude Fable 5 announced — most capable widely-released model, always-on adaptive thinking, 1M-token context window, and 128K output tokens. Note: access was suspended June 12 due to a US government export-control directive.

"Code with Claude: Extended Tokyo" event for independent developers and early-stage founders — Boris also noted in a recent interview that he hasn't written a line of code himself in 2026, living in the agentic future he built.

"A harness for every task: dynamic workflows in Claude Code" — Thariq wrote up how Claude can now write its own orchestration harness on the fly, custom-built per task rather than using a fixed workflow template.