Sometimes the most telling release note is the one with nothing spectacular in it. Claude Code 2.1.220 shipped on July 25, a day after the big 2.1.219. The entire changelog entry reads: «Bug fixes and reliability improvements». That’s it.
The context is what makes it interesting
To see why this is worth a mention, you need to know the version before it. 2.1.219 was a heavy one: Opus 5 became the new default Opus model, with 1M context and fast mode at $10/$50 per million tokens. On top of that came a new sandbox.network.strictAllowlist setting, the DirectoryAdded hook, nested subagents up to depth 3 by default, and a dozen more changes. A big release, right in the middle of the Opus 5 launch.
And that’s exactly when bugs creep in. New model rows in the /model picker, changed fast-mode logic, adjusted subagent depth — those are the spots where edges show. 2.1.220 is the answer: no new features, just stability and reliability.
Why this isn’t a boring update
A cleanup release right behind a feature release is a sign of maturity. It says the team isn’t just pushing the next big thing — it’s checking that the last one actually runs clean. If you use Claude Code daily on real projects, you don’t notice releases like this through new buttons. You notice them because something that snagged yesterday just works today.
Update advice: run claude update. If you’re already on 2.1.219 with Opus 5, this mostly buys you a calmer foundation.
My take
I like this kind of release. The AI world is stuck in permanent announcement mode — a new record, a new model, a bigger context, every single day. So an update that honestly says «we just fixed bugs» is almost refreshing. Good tools aren’t built from features alone, but from the unspectacular days in between. 2.1.220 is one of those days.
Sources: