Sonnet 5 Becomes the Default, Background Subagents, and Export Controls Lift

I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of July sixth. Claude Sonnet five is now the default in Claude Code. Subagents now run in the background by default. And the U.S. lifted export restrictions on Anthropic's frontier models.

Let's start with Claude Code -- the headline this week is that Claude Sonnet five is now the default model, and it brings a native one million token context window along with promotional pricing of two dollars and ten dollars per million tokens through August thirty-first. Beyond the raw context bump, there are quality-of-life changes bundled in. Subagents and context compaction now inherit your session's extended thinking configuration, so delegated work should feel more consistent with your main conversation. Sonnet five sessions also skip a mid-conversation system role that used to inject harness reminders, which means cleaner context throughout.

That shift toward cleaner, less interruptive behavior shows up in the subagent changes too. Subagents now run in the background by default, so Claude keeps working while they execute and notifies you when they finish, rather than blocking the turn. Background sessions launched from the agents view will also commit, push, and open a draft pull request when the code work is done -- no more stopping to ask. Session rows show those pull request links inline.

On the interaction side, prompts that ask you a question now wait for your input by default instead of timing out and continuing on their own. If you liked the old auto-continue behavior, you can opt back into an idle timeout through your config. And a small but useful clarity change -- the permission mode formerly called default is now labeled Manual across the CLI, Visual Studio Code, and JetBrains. The old name still works, but the new one better describes what it actually does.

Claude in Chrome also reached general availability this week, with plan mode now correctly prompting for state-changing browser calls while auto-allowing read-only ones. And for teams, admins can now set an organization-wide default model from the console, which shows up in your model picker as Org default or Role default when you haven't picked one yourself.

Moving to the under-the-hood work -- this week's fixes mostly cluster around background agent stability and network resilience. On the stability side, a batch of fixes stops the Linux daemon from killing itself roughly every fifty seconds after an unclean shutdown, prevents the stop command from being undone by respawns, and keeps older reinstalled builds from hijacking the daemon. Sessions that were silently stalling after sleep or wake now resume properly, and macOS users should no longer see that Reconnecting message every fifty-two seconds in the agents view. The network improvements follow the same theme of not giving up too easily.

Brief network drops mid-response now retry with backoff instead of aborting your turn, transient rate limit errors unrelated to your usage cap are retried automatically for subscribers, and mid-stream server errors preserve the partial output with an incomplete-response notice. TLS certificate errors go the other direction -- they now fail immediately with actionable guidance instead of burning through retries. And one performance win worth noting -- the code review command cuts token usage by roughly twenty-five percent by merging five cleanup finders into one.

On the broader Anthropic news side, the U.S. story is the one to lead with. The New York Times is reporting that the Trump administration has lifted export controls on Mythos and Fable, resolving a months-long dispute that had gated access to Anthropic's frontier models. If your team's model access was affected, this is the formal end of that restriction. That easing on one side is paired with tension on another.

Reuters is reporting that Alibaba is banning internal use of Claude Code over alleged backdoor concerns, the latest escalation in the ongoing dispute between Anthropic and Alibaba. It's useful context if you follow how model-provider tensions ripple into API and tooling access. And looking further out, The Information reports that Anthropic is in talks with Samsung to manufacture a custom AI chip, a move that could affect long-term inference cost and capacity for Claude Code workloads. We'll link all three stories in the show notes, along with the Sonnet five announcement and a new workbench called Claude Science for research workflows.

That's it for the brief. I'm Shannon, and we'll see you next week.