WHY users should care
Claude Opus 5 is live — if you're building serious agentic stuff, the default Opus model just got a 1M context window, adaptive thinking, and a May 2026 knowledge cutoff. It's the new recommended starting point for complex enterprise/agentic work.
Subagent depth doubled — multi-agent pipelines can now go 3 levels deep by default (was 1). If you're building workflows that spawn agents, this is a structural capability unlock.
Key changes (v2.1.219)
- 🚀 Claude Opus 5 default — `claude-opus-5`, 1M context, adaptive thinking on; $10/$50 MTok input/output on Anthropic API / Bedrock / Vertex
- 🚀 Nested subagent depth 3 — sub-sub-agents now work out of the box; disable with `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1`
- `sandbox.network.strictAllowlist` — deny non-allowlisted hosts without prompting (CI/headless hardening)
- `DirectoryAdded` hook — fires after `/add-dir` registers a new working dir mid-session
- Dynamic workflow size — defaults to medium (<15 agents); configurable via `workflowSizeGuideline` in settings
- Two new API beta headers — `server-side-fallback-2026-07-01` + `fallback-credit-2026-07-01`
- `mcp_server_errors` in headless init — broken MCP servers surfaced at startup now
- ⚠️ Opus 4.1 deprecating — `claude-opus-4-1-20250805` retires 2026-08-05; migrate to Opus 5
- Opus 4.7 dropped from fast mode (fast mode now = Opus 5 + Opus 4.8)
Reference
Docs PR: thevibeworks/claude-code-docs#1091
WHY users should care
Claude Opus 5 is live — if you're building serious agentic stuff, the default Opus model just got a 1M context window, adaptive thinking, and a May 2026 knowledge cutoff. It's the new recommended starting point for complex enterprise/agentic work.
Subagent depth doubled — multi-agent pipelines can now go 3 levels deep by default (was 1). If you're building workflows that spawn agents, this is a structural capability unlock.
Key changes (v2.1.219)
Reference
Docs PR: thevibeworks/claude-code-docs#1091