Opus 4.7 Arrives With a New Effort Level, Native Binary, and Parallel Code Review

I'm Shannon, and this is the Claude Notes Brief -- your weekly rundown of Claude Code updates and Anthropic news for the week of April twentieth. Opus four point seven is now available in Claude Code with a new effort level. Anthropic is fielding investor interest at eight hundred billion dollars. And The Information reports on Claude Code's enterprise costs.

Nine releases for Claude Code this week, headlined by Opus four point seven and a new effort level called extra-high that sits between high and max. If you've been toggling between high effort for speed and max effort for thoroughness, extra-high gives you a middle ground -- more intelligence without the full latency hit of max. There's also a new interactive slider you can pull up to dial in exactly where you want that speed-intelligence tradeoff to land. And if you're a Max subscriber, auto mode now works with Opus four point seven out of the box, no special flag required.

Anthropic published a companion blog post covering best practices for the new model, including how the effort levels have been recalibrated -- we'll link it in the show notes. That model update is the headline, but the infrastructure change underneath it matters just as much. Claude Code now ships as a native binary instead of running bundled JavaScript through Node. Your commands and flags all work the same way, but startup should feel noticeably snappier.

It's the kind of change you notice most when you're firing off quick one-line prompts throughout the day. On the workflow side, there's a new parallel code review command called ultra-review. You run it, and it launches multiple agents in the cloud to analyze your current branch at the same time. You can also point it at a specific GitHub pull request number.

Because the review runs in the cloud, it doesn't block your local session -- you can keep working while the review finishes. The theme of reducing friction extends to permissions too. A new command scans your recent transcripts and proposes an allowlist of read-only commands you've been approving repeatedly, then writes it to your settings file. On top of that, read-only commands with glob patterns and commands prefixed with a directory change into your project folder no longer trigger prompts at all.

One more worth mentioning -- session recaps. When you come back to a session after stepping away, Claude Code now shows a summary of what happened while you were gone. You can also invoke it manually anytime. It works even with telemetry disabled, and you can opt out entirely through your config if you prefer.

Under the hood, the security and stability work this week was substantial. Bash permission rules got a serious hardening pass. Deny rules now catch commands wrapped in things like sudo, env, and watch -- previously those wrappers could slip past your rules. Allow rules for the find command no longer auto-approve dangerous flags like execute or delete.

On macOS, symlinked system paths are now treated as dangerous targets for removal. And a sandbox bypass that allowed permission prompts to be skipped entirely has been fixed. That security work pairs with a round of terminal rendering fixes. If you use iTerm2 with tmux, Ghostty, Kitty, Alacritty, or WezTerm -- especially over SSH or mosh -- you should see fewer display glitches.

Tearing, garbled output, and washed-out colors have all been addressed. Memory usage also dropped, thanks to on-demand loading of language grammars for syntax highlighting. And there's a new fullscreen mode for the terminal interface that eliminates flicker, along with readline-style keybindings for jumping to the start or end of a line and deleting words.

Beyond Claude Code, a few stories worth tracking this week. Bloomberg is reporting that Anthropic is fielding investor offers at a valuation of up to eight hundred billion dollars. That number signals sustained confidence in the company's trajectory and, practically speaking, it means continued investment in the infrastructure behind Claude Code and the broader model lineup. On the policy side, the New York Times reports that Anthropic CEO Dario Amodei met with White House officials to discuss government access to the Mythos model.

The meeting was described as productive, with both sides aiming for a compromise. If your organization depends on Anthropic's API, the outcome of those negotiations could eventually shape model availability and compliance requirements. And on the enterprise front, two stories from The Information paint a picture of Claude Code's growing footprint and the costs that come with it. One piece covers how Uber's engineering team encountered unexpectedly high costs running Claude Code at scale -- a useful data point if you're planning a broad rollout.

The other reports that Anthropic is shifting enterprise customers to usage-based pricing as demand increases, which could change how teams budget for Claude Code going forward. That's it for the brief. I'm Shannon, and we'll see you next week.