Claude Opus 5 in Claude Code: What Changed for Developers

The announcement

On July 24, 2026, Anthropic published "Introducing Claude Opus 5," describing a new Opus-tier model built to push further on long-running agents, coding, and other professional work. The post positioned Opus 5 as the top of Anthropic's current lineup, a model reserved for the hardest tasks rather than a routine swap-in for whatever ran before it.

Claude Code's own release notes caught up within days. The tool's "What's New" digest for Week 30 (July 20 to 24, 2026, spanning versions v2.1.214 through v2.1.219) documented the practical side of that announcement: Claude Opus 5 became the new default Opus model inside Claude Code itself, alongside two concrete specs that matter more to a working developer than a launch post's framing usually does.

What actually changed

Two numbers anchor the update. Claude Opus 5 in Claude Code ships with a 1M-token context window, and it adds a fast mode priced at $10 per million input tokens and $50 per million output tokens. The context window matters for anyone who has watched a long agent session hit a wall mid-task, getting compacted and losing earlier reasoning or file state in the process. A larger window means more of a real codebase, more of a long conversation, and more prior tool output can stay in view without Claude Code trimming it for you.

The fast mode is the other half of the change, and it may be the more practical one day to day. Instead of paying full Opus-tier rates for every request in a session, fast mode gives Claude Code a lower-cost, lower-latency option for the lighter parts of a workflow — quick lookups, short edits, scaffolding — while still running on the same underlying model. Selecting which model backs an active session, including switching to Opus 5 mid-project, is an existing Claude Code capability; the session below shows the general shape of that kind of switch.

Also shipped in the same Week 30 window, worth knowing even though this piece stays focused on Opus 5: a public-beta iOS Simulator pane in Claude Code Desktop that lets Claude run your app and tap through it while you watch, a Claude Security plugin that runs a multi-agent vulnerability scan of a codebase and turns the findings you select into patches you apply yourself, and a /code-review command that now runs as a background subagent instead of blocking the terminal while it works.

Who should care

The clearest beneficiaries are developers running long agent sessions against large codebases — the kind of work where the old context ceiling forced Claude Code to summarize or drop earlier context partway through a task. If your sessions routinely involve multi-file refactors, extended debugging, or agents that need to remember decisions made an hour earlier, the 1M-token window is the change worth testing first.

It also matters for teams that already made Claude Sonnet 5 their default. Sonnet 5 picked up its own native 1M-token context window and adaptive thinking turned on by default several weeks earlier, in Claude Code's Week 27 release, and has been doing the bulk of everyday work since. Opus 5 is not a replacement for that default; it is the escalation path Anthropic built above it, for architecture decisions and coding problems where the extra capability is worth the extra cost. Developers doing routine edits may not notice much difference reaching for Sonnet 5 instead.

Availability and how to try it

Because Opus 5 is now the default Opus model in Claude Code, most developers on a plan with Opus-tier access will pick it up automatically the next time they update. Session-level model selection already exists in Claude Code, so switching a specific session to Opus 5 — or back to Sonnet 5 for lighter work — doesn't require anything exotic.

Before changing a default for a whole team, check three things: that your Claude Code build falls in or after the Week 30 version range (v2.1.214 to v2.1.219), that your plan's billing actually covers Opus-tier and fast-mode pricing, and that the official Week 30 docs page still describes the same behavior — release notes get amended, and it is worth reading the current version rather than trusting a summary, including this one.

What to watch next

Two things are worth tracking: whether fast-mode pricing or the context ceiling move again in a future weekly digest, and whether Anthropic narrows the practical gap between Sonnet 5 and Opus 5 now that both carry 1M-token context windows. The distinction Anthropic is drawing right now is capability tier, not context size, and that framing could shift as usage data comes in.

For now, treat this as a model-selection decision rather than a mandatory upgrade. Verify the Week 30 notes and the Opus 5 announcement directly before wiring a new default into CI configs, agent pipelines, or anything that runs unattended, and re-check before assuming a colleague's account behaves the same way yours does.