Opus 4.8 Becomes the Cloud Default, and a Nineteen Billion Dollar Lease

I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of July thirteenth. Claude Opus four point eight is now the default on cloud providers. The doctor command can fix issues, not just report them. And Ben Bernanke joins Anthropic's oversight trust.

Let's start with Claude Code, where the headline this week is that Claude Opus four point eight is now the default model on Bedrock, Vertex AI, and Claude Platform. If you run Claude Code through any of those cloud providers, your sessions will pick up a different baseline for cost, latency, and output quality going forward. You can still switch models from the command line if you need to, but the new default is Opus four point eight. That model change lines up with another shift on the same providers -- auto mode is now generally available on Bedrock, Vertex, and Foundry, so you no longer need the opt-in environment variable to turn it on.

If your organization needs it off, there's a settings flag for that, and one thing worth flagging: the auto mode setting is no longer read from the local project settings file. You'll want to move it up to your user-level settings. Alongside those cloud changes, the doctor command got a real upgrade. It used to just report installation and configuration issues -- now it actively diagnoses them and offers to fix them for you.

It'll also suggest trimming your project's Claude memory file by cutting content Claude could figure out from the codebase on its own, and it will tell you if your Claude launcher is being externally managed by something like a package manager. Background agents also got smarter lifecycle handling. When Claude Code updates, background agents now upgrade in the background right away, so attaching to a stale session doesn't cost you a slow upgrade. You'll also get a warning before your login expires, and the agent view now links to any pull requests a session touched -- edited, merged, commented on, or pushed to.

And two smaller workflow wins to round things out. The commit-push-pr command now respects your configured push remote, so fork-based workflows and repos where origin isn't your push target won't prompt you anymore. And the review command has reverted to a fast single-pass review -- if you want the multi-agent version at a chosen effort level, that now lives under code-review, which the team says produces better findings on Opus four point eight.

On to under the hood, and the theme this week is responsiveness. A handful of terminal issues got resolved: the freeze you'd sometimes see on very long streamed responses, transcript jumping when a response finished, content jumping while you were scrolling through long transcripts, and resume and continue sessions not responding to keyboard input on startup. Interactive sessions also stop re-analyzing the entire transcript after every turn, which cuts per-turn CPU and memory use noticeably on longer sessions. There's also a security fix worth mentioning.

User config references in shell-form plugin hooks, monitors, and MCP header helpers are now rejected to prevent shell injection. If you were using that pattern, you'll want to switch to the exec form or the environment variable equivalent. And one more memory improvement -- the auto-updater now streams downloads to disk instead of holding them in memory, which cuts about four hundred megabytes off peak usage during updates.

Now to the broader Anthropic news. The Wall Street Journal is reporting that Chinese authorities have publicly alleged security vulnerabilities in Claude Code, escalating an ongoing dispute over access and trust in the tool. If your organization operates across jurisdictions, or you've been tracking the earlier Alibaba story, this is worth following. On the business side, Wired has a piece on how Anthropic plans to price Claude Fable five for consumers, including how access and cost will differ from previous models -- useful context if you're planning Claude Code budgets.

And two governance and infrastructure items to close out. Bloomberg reports that former Federal Reserve chairman Ben Bernanke has joined Anthropic's Long-Term Benefit Trust, the body that shapes governance decisions at the company. And the Wall Street Journal has the details on a nineteen billion dollar lease Anthropic signed with TeraWulf for an AI infrastructure campus in Kentucky -- one of the more concrete signs of the compute buildout behind the models you use every day. We'll link all of those in the show notes.

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