Shipped.

Anthropic confirmed via subscriber email and Help Center on June 15: the planned Agent SDK / claude -p / third-party-app billing split is no longer happening. Subscription limits apply as before. Previously announced May 14, the change would have moved programmatic usage to a separate monthly credit ($20 for Pro, up to $200 for Enterprise). Now on hold indefinitely. Reported by The Decoder, citing Anthropic’s Help Center and subscriber communications.

Follow-up bug-fix release. Preserves partial responses on mid-stream connection drops instead of surfacing raw errors. Fixes mouse-wheel scrolling regression in WSL2 under Windows Terminal and VS Code (regression since 2.1.172), Linux sandbox glob crash over large directories, feedback survey capturing single-digit replies as session ratings, welcome screen stacking multiple promo banners, Ctrl+O not showing subagent transcript, prompt-input focus loss when subagent and footer panels are visible, remote background tasks appearing stuck as “still running,” and plugin loading performance in remote sessions.

Privacy-preserving analysis of approximately 400,000 Claude Code sessions, October 2025 through April 2026. In a typical session: people drive planning decisions and Claude drives execution decisions. More expert users get more Claude output per instruction. Non-engineers succeed at coding tasks at nearly the same rate as software engineers. GitHub projects with coding agent activity have more than doubled since late 2025. The average Claude Code user now logs 20 hours per week in sessions.