On November 2, 2026, Anthropic opened a Message Threads beta on the Claude Messages API. Instead of sending the full conversation history with every request, apps send only new messages while the server holds the context β cutting median time to first token by roughly 35%.
Anthropic added the claude plugin eval command to Claude Code v2.1.269 (September 11, 2026). Every test case runs twice β once with the plugin loaded (with-arm) and once without (without-arm) β and the difference, Ξ, is the plugin's measured contribution. Six grader types, JSON + HTML reports, and CI threshold flags complete a plugin quality-assurance pipeline.
Since Claude Code v2.1.224, separate sessions can message each other directly. ListAgents discovers reachable sessions; SendMessage delivers text to one by name. When one session's change breaks what another is building, it can warn that session before you notice β reshaping how parallel agent development coordinates.
Anthropic released Claude Fable 5.1 on September 1, 2026. At the same price as Fable 5 ($10/$50), cache read costs dropped 75% β from $1.00 to $0.25 per million tokens. Per-message effort control, now in beta, lets you change the reasoning budget mid-conversation without invalidating the prompt cache. 1M token context and always-on adaptive thinking are standard.
In April 2026, Anthropic released ant β an official CLI for the Claude API. Build requests with YAML instead of curl, natively integrate with Claude Code, and manage agents, sessions, and deployments directly from your terminal. An analysis of how AI app development workflows are changing.