Claude Sonnet 5 in Claude Code - What Actually Changed for Your Business

The short answer

Claude Sonnet 5 became Claude Code's default model on June 30, 2026, replacing Sonnet 4.6 with no action needed on your end - same desktop app, same plan, same conversation window. The one change worth knowing about is that adaptive thinking is now on by default, which means Claude decides for itself how much extra reasoning a step deserves before it acts, instead of moving at the same speed on every task regardless of difficulty. Rate limits moved up at the same time to make room for that extra thinking, so a longer build session has more headroom than it did before the switch, not less.

What actually changed on June 30

Anthropic shipped Claude Sonnet 5 as a drop-in replacement for Sonnet 4.6 across every Claude plan, including Claude Code. Nothing about the interface moved - the desktop app looks the same, your existing CLAUDE.md files and skills still work the same way, and you did not have to opt in to anything. The model swapped underneath the product you were already using.

The bigger practical change is adaptive thinking. Earlier Claude models spent roughly the same amount of reasoning on every step of a task, whether that step was trivial or genuinely hard to get right. Sonnet 5 decides for itself, mid-task, when a step is worth slowing down for - checking a piece of logic twice, working through an edge case, re-reading a file before editing it - and moves quickly through the steps that do not need that.

Why this matters if you do not write code

The failure mode I coach members through more than any other is Claude finishing a build that looks done and is not - a form that submits without validating an email address, a report that pulls the wrong date range, a script that silently skips a row it could not parse. That gap between 'looks finished' and 'is actually correct' is usually a place where the model needed to slow down and check something, and moved past it instead. Adaptive thinking is aimed directly at that gap: the model is now built to notice when a step needs a second look, not just to produce an answer as fast as possible.

This does not replace your own review of the finished build - nothing does, and no version of this model changes that. What it does is shrink the number of avoidable mistakes that show up because the model rushed a step it should have double-checked.

The rate limit increase - more room, not less

Rate limits went up across Chat, Cowork, Claude Code, and the Claude Platform at the same time Sonnet 5 launched, specifically because adaptive thinking uses more tokens on the steps where it decides to think harder. If you had hit Claude Code's usage ceiling mid-project before, it is worth checking again - the ceiling itself moved, so a build that used to run out of room partway through may not anymore.

How to actually notice it working

Run the Pause Test. Give Claude Code a task with one deliberately awkward edge case buried in it - a date format that could be read two ways, a discount rule with an exception, a field that could be blank. Watch what happens before it writes anything: does it flag the ambiguity in its own plan, or does it plow through and hope? That is the Pause Test, and it is the simplest way to feel the difference between an assistant that types fast and one that decided the answer was worth getting right first.

[How to use Claude Code for your business](/blog/how-to-use-claude-code-for-your-business) is the right starting point if this is your first real build. Once you are past that, [how to give Claude Code memory of your business](/blog/how-to-give-claude-code-memory-of-your-business) covers the CLAUDE.md method that compounds with adaptive thinking - the more accurate context Claude has about how your business actually works, the more useful its extra reasoning becomes.

What to actually do about it

1. Do nothing to opt in - if you are on Claude Code already, you are on Sonnet 5. There is no toggle to find.
2. Re-test any build that used to hit a rate limit mid-session, since the ceiling moved along with the model.
3. Feed it the hardest part of a project first, not last, since that is where adaptive thinking changes the outcome the most.
4. Keep reviewing the finished work the way you already do. A smarter default model changes how often it catches its own mistakes - it does not remove your job of checking the result.

Where this fits

[How to use Claude Code subagents for your business](/blog/how-to-use-claude-code-subagents-for-your-business) is worth pairing with this - subagents split a project into pieces you can delegate, and each piece now runs on a model that thinks harder about the parts that need it. The two upgrades stack.