Anthropic · 2026-07-24 · major
Claude Code 2.1.219 — Opus 5 becomes default, subagents nest to depth 3
Claude Code 2.1.219 makes Claude Opus 5 the default Opus model in the CLI, adds a 1M-token context in fast mode, subagent nesting up to depth 3, and a strict sandbox network allowlist that refuses non-allowed hosts without prompting.
Anthropic's terminal coding agent adopts Opus 5 as its default and lets subagents spin up their own subagents three layers deep.
Quick facts
| Maker | Anthropic |
|---|---|
| Version | 2.1.219 |
| Released | 2026-07-24 |
| New default Opus | Claude Opus 5 (1M context) |
| Fast mode price | $10 in / $50 out per 1M tokens |
| Subagent nesting | 3 levels deep (was 1) |
| Removed from fast mode | Opus 4.7 |
Pricing
| Opus 5 fast mode — input | $10.00 / 1M tokens |
|---|---|
| Opus 5 fast mode — output | $50.00 / 1M tokens |
What is it?
Opus 5 replaces Opus 4.7 as Claude Code's default Opus model in v2.1.219, arriving with a 1M-token context and a fast-mode price of $10 in and $50 out per million tokens. The release also adds a strict sandbox network allowlist that refuses non-allowed hosts without prompting the user, plus a new DirectoryAdded hook that fires when `/add-dir` or the SDK's `register_repo_root` request registers a working directory mid-session.
How does it work?
Subagents spawned by Claude Code can now spawn their own subagents down to three levels deep — the previous limit was one — with `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1` turning nesting off for workflows that assume a flat structure. Nested subagents at depth 2 or more surface in the headless stream-json output when `--forward-subagent-text` is set, keyed by their spawning tool_use id, so an orchestrator can see the whole tree instead of only its immediate children.
Why does it matter?
Users of the Claude Code CLI pick up Opus 5's stronger long-horizon coding without changing model flags, and the deeper subagent tree lets one automated workflow decompose a task into workers of workers rather than a flat fan-out — the shape most complex orchestration actually wants. The stricter sandbox allowlist also removes a common approval-fatigue paper cut when running untrusted commands in headless CI or a batch job.
Who is it for?
Claude Code power users, teams running Claude Code in CI, and anyone building multi-agent workflows on top of the CLI.
Frequently asked questions
- What changed for Opus 4.7 users in Claude Code 2.1.219?
- Claude Code 2.1.219 removes Opus 4.7 from fast mode. The `/fast` flag now routes to Opus 5 or Opus 4.8 only, so anyone who was pinning `/fast` to Opus 4.7 in the CLI must pick a different model or drop out of fast mode entirely.
- How does deeper subagent nesting work in Claude Code 2.1.219?
- Claude Code 2.1.219 lets a subagent spawn its own subagents down to three levels deep, up from a hard cap of one. Set the `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1` environment variable to keep the old flat behaviour when a workflow assumes a single fan-out.
- What is the strict sandbox allowlist in Claude Code 2.1.219?
- A new `sandbox.network.strictAllowlist` setting causes the Claude Code sandbox to refuse any host that isn't on the explicit allowlist instead of prompting for approval. Operators can lock down which domains a sandboxed command may reach without a human having to click through every unknown request.
- How much does Claude Opus 5 cost in Claude Code fast mode?
- Claude Opus 5 in Claude Code's fast mode is billed at $10 per million input tokens and $50 per million output tokens, per the v2.1.219 release notes. Standard-mode pricing follows the usual Anthropic API rates rather than the fast-mode surcharge.
Try it
npm install -g @anthropic-ai/claude-code@2.1.219