Sonnet 4.6 Launches, Pentagon AI Dispute, and Claude Code's Market Impact

I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of February twenty-second. Claude Sonnet four point six lands with a one million token context window. The Pentagon threatens to cut ties with Anthropic over safety restrictions. Bloomberg traces Claude Code's unlikely rise.

Let's start with Claude Code -- five releases this week, and the throughline is massive context windows paired with better isolation and workflow control. The headline is Claude Sonnet four point six, now available and replacing Sonnet four point five on the Max plan. If you're on Max, you'll want to switch models to get the upgrade. Sonnet four point six brings the full one million token context window to the mid-tier model -- that's the same massive context that was previously exclusive to Opus.

Fast mode on Opus four point six now also gets one million tokens of context. That focus on expanded context connects to the new git worktree isolation feature. You can now launch Claude in an isolated git worktree using the dash w flag, keeping experimental work completely separate from your main branch. Individual agents can also declare worktree isolation in their configuration to always run in their own isolated environment -- perfect for risky operations or testing workflows.

The workflow improvements extend to Visual Studio Code too, where plan previews now update in real time as Claude works through iterations. No more manual refreshing -- you see progress live. If you reject a plan, the preview stays open so Claude can revise it right in place. Simple mode got a major upgrade this week that makes it truly minimal.

When you enable simple mode, Claude now strips out MCP tools, attachments, hooks, and even Claude dot MD loading entirely -- giving you a predictable, bare-bones environment. But it's not too bare -- the file edit tool is now included alongside Bash, so you can still make direct file changes when needed.

Under the hood, this was a week focused on memory and performance across the board. The team shipped extensive memory leak fixes covering agent task state, LSP diagnostics, shell commands, and circular buffers. Those fixes are paired with startup improvements that cut time-to-interactive by around five hundred milliseconds in headless mode and reduce redundant network calls during MCP initialization. Several reliability issues got attention too.

Plan mode and custom session names now survive context compaction -- no more losing your work when Claude hits memory limits. Sessions with large first prompts over sixteen kilobytes no longer disappear from the resume command. And on Windows, hooks now execute correctly via Git Bash, fixing a platform-specific bug that was breaking automation workflows.

Beyond Claude Code, it was a week that highlighted just how central the tool has become to Anthropic's story. Bloomberg published a deep dive tracing how Claude Code -- launched roughly a year ago -- became the surprise hit that transformed Anthropic into what they call an AI juggernaut. The piece details how the coding tool pushed rivals to accelerate their own development environments and became the product that defined Anthropic's trajectory. That success story comes against a backdrop of growing tensions with the Pentagon.

The New York Times reports on an escalating dispute between Anthropic and the Defense Department over AI safety restrictions on Claude. The conflict centers on usage policies that directly affect the models powering Claude Code -- the same restrictions that govern how you can use the tool. Meanwhile, the market is taking notice of Claude Code's expanding reach. Bloomberg covered how cybersecurity stocks slid Friday after Anthropic unveiled Claude Code Security, with investors apparently spooked by the implications of AI-powered security tooling.

It's a signal of how seriously the industry is taking Claude Code's growing capabilities. That's it for the brief. I'm Shannon, and we'll see you next week.