TerminalBlog

Posted on Jul 31 • Originally published at terminalblog.com

Claude Code Gets Opus 5 as the Default — What the 1M-Context Model Changes for You

#claudecode #guide #claudeopus5 #release

On July 24, 2026, Anthropic shipped Claude Code v2.1.219 — and it is not a typical patch release. This version makes Claude Opus 5 the default Opus model inside Claude Code, meaning every user on a Max, Team Premium, or API plan now gets Opus 5 without changing a single setting. If you have been waiting for a reason to try Claude Code, or wondering whether to update, this is the version that changes the math.

This guide walks through what Opus 5 actually is, what changes when you update, and which gotchas matter if you are new to Claude Code or coding agents in general.

What is Claude Opus 5?

Claude Opus 5 is Anthropic's most capable model. It replaces Opus 4.8 as the top-tier option inside Claude Code. The headline specs:

1 million token context window — both the default and the maximum. No smaller variant exists. You get the full window without asking for it.
128K max output tokens — the model can write very long responses without hitting the output ceiling.
Thinking on by default — the model decides when and how deeply to reason on each turn. You no longer need to enable thinking manually.
Fast mode — a research-preview speed tier priced at $10/$50 per million tokens (input/output), available on the Claude API.
Pricing unchanged — $5/$25 per million tokens (input/output), same as Opus 4.8.

Compared to Opus 4.8, Anthropic describes Opus 5 as a "step-change improvement" rather than an incremental update. The biggest gains are in deep reasoning, agentic coding (long multi-file tasks), test-time compute scaling (turning up the "effort" dial actually produces better results), and code review accuracy.

What changed in Claude Code v2.1.219

The v2.1.219 release brings several practical changes beyond the model swap. Here is what matters for everyday use:

1. Opus 5 is now the default

If you type claude and press Enter, the model picker now defaults to Opus 5 for Anthropic API, Max, Team Premium, and Enterprise pay-as-you-go users. Pro and Team Standard users default to Sonnet 5 instead. If you previously had Opus 4.8 selected, you will see the upgrade automatically.

To check your current model, run /model inside Claude Code. The picker now highlights the newest model name to make it clear which version you are on.

2. Thinking is on by default

This is the most important behavioral change for beginners. On Opus 4.8, Claude Code ran without thinking unless you explicitly enabled it. On Opus 5, the model thinks by default — it decides on each turn whether and how deeply to reason before responding.

For most users, this means better answers with zero configuration. The model handles the thinking budget automatically. If you want more control, you can adjust the effort level: low, medium, high (the default), xhigh, or max. Higher effort means deeper reasoning at the cost of more tokens and time.

3. Nested subagents go three levels deep

Before v2.1.219, Claude Code subagents could spawn only one level deep (depth 1). Now subagents can spawn their own subagents up to depth 3 by default. This means Claude can delegate a task to a subagent, which can delegate a subtask to its own subagent, which can delegate once more.

If you have a complex task like "refactor the authentication module, update the tests, and verify the API endpoints still pass," Claude can now parallelize more aggressively by spinning up a tree of workers. To disable nesting entirely, set:

export CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1

4. Dynamic workflow size guidelines

Claude Code now defaults to a medium dynamic workflow size — aiming for fewer than 15 agents in a workflow at a time. You can change this in /config under "Dynamic workflow size" or set it from any settings file using the new workflowSizeGuideline key. The current default size is also shown in the running-workflow status line.

5. Sandbox network allowlist

A new sandbox.network.strictAllowlist setting lets you deny all network connections to hosts not on your allowlist — without being prompted each time. This is useful if you want tighter security in sandboxed environments and do not want Claude Code to ask permission for every outbound connection.

6. DirectoryAdded hook

A new DirectoryAdded hook fires whenever you add a working directory mid-session (via /add-dir or the SDK's register_repo_root). This lets you react to directory changes with custom logic — for example, automatically scanning new directories for linting issues or setting up project-specific context.

Gotchas for beginners

Opus 5 introduces a few breaking changes and quirks that catch people off guard:

Disabling thinking requires effort "high" or below. If you set thinking to disabled while effort is xhigh or max, you will get a 400 error. This is a breaking change from Opus 4.8, where thinking and effort were independent. Either lower the effort level or leave thinking enabled.

Opus 4.7 is removed from fast mode. The /fast command now only applies to Opus 5 and Opus 4.8. If you were using fast mode with Opus 4.7, it no longer works.

Remove old verification instructions. Opus 5 verifies its own work without being told to. If you have prompt templates that say "include a final verification step" or "use a subagent to verify," remove them — they cause over-verification on Opus 5 and waste tokens.

Longer default responses. Opus 5 tends to write longer, more detailed responses by default. If you prefer concise output, lower the effort level or add explicit length instructions to your prompt.

How to update

If you installed Claude Code via npm:

npm update -g @anthropic-ai/claude-code

Or run the update command inside Claude Code:

claude update

After updating, run claude --version to confirm you are on v2.1.219 or later. Then run /model to see that Opus 5 is now your default.

What this means for coding agent users

The Opus 5 default in Claude Code raises the baseline for what a terminal coding agent can do. The 1M context window means Claude can read your entire repository without losing track of earlier instructions. The thinking-on-by-default approach means fewer blind spots in complex refactors. And nested subagents at depth 3 mean Claude can break large tasks into parallel work streams without your manual orchestration.

If you are choosing between Claude Code and other terminal agents (Codex, Copilot CLI, OpenCode), Opus 5 widens the capability gap for reasoning-heavy tasks. Copilot CLI also added Claude Opus 5 support in v1.0.75 (released the same day), but Claude Code is the first to make it the default with deep integration — thinking controls, effort tuning, fast mode, and nested subagent depth all built around the model's strengths.

The bottom line: update to v2.1.219, accept the Opus 5 default, and let the model think. Most users will see better results without changing anything.

DEV Community