
Fable 5.2 Becomes the Default, and Code Mode Arrives
Claude Fable 5.2 is now the default model with a one million token context window, and an experimental Code Mode lets Claude write JavaScript to invoke its own tools. Also this week: sandboxed Bash commands now prompt for permission, the Visual Studio Code extension gains prompt suggestions and live command output, and Anthropic faces scrutiny over spending plans and a lawsuit seeking to halt frontier model development.
Chapters
Transcript
I'm Shannon, and this is the Claude Notes Brief -- Claude Code updates and Anthropic news for the week of Monday, November sixteenth. Claude Fable five point two is now the default model. An experimental mode lets Claude write JavaScript to call its own tools. And Anthropic's spending plans are drawing scrutiny.
Let's start with Claude Code, where Claude Fable five point two is now the default model, and it brings a one million token context window with it. Pricing is eight dollars per million input tokens and forty dollars per million output, with cache reads at twenty cents per million. If you had Fable five point one saved as your pick, that selection moves over to the new default automatically on Pro and Max plans, and on Team and Enterprise once your organization allows it. You can still select the older version from the command line if you'd rather stay put.
Alongside the model change, there's a new experimental feature called Code Mode. It's opt-in, and the idea is that instead of issuing individual tool calls one at a time, Claude writes JavaScript to invoke its own tools. On multi-step work, that can cut down the number of round trips considerably. The documentation covers how to turn it on, and how administrators can disable it across an organization -- we'll link it in the show notes.
The context window change has a second effect worth flagging. If you point Claude Code at a gateway or a third-party model it doesn't recognize, it now assumes one million tokens of context instead of two hundred thousand. On a model with a smaller real window, that can leave a session failing every turn once it fills up, so there's an environment setting to disable the one million token assumption, or you can compact earlier. Permissions also shifted.
If you run with the Bash sandbox turned on, sandboxed commands no longer skip the permission prompt by default. There's a setting to restore the old behavior, and if your sandbox is enabled through managed settings, that key needs to live in your highest-ranking managed source. On the editor side, the Visual Studio Code extension now shows the same prompt suggestions you get in the terminal -- Tab to accept, and a setting to turn them off. It also streams live output from running Bash or PowerShell commands underneath the command card, and you'll see a trust prompt before your first session in an untrusted folder.
Two smaller items round things out: fast mode now works through Vertex in the global, US, and EU regions, falling back to standard speed when Vertex declines. And the cap on how many messages your sessions on the same machine can send each other without your involvement went from about twenty to one hundred thousand, so multi-session workflows no longer stall partway through.
Under the hood, the theme this week is closing permission gaps and trimming memory. A security issue in how zstd-compressed responses are decoded has been patched, and it ships alongside a batch of fixes where compound commands, sudo and env wrappers, wildcard allow rules, and copy and move targets could all skip a permission prompt. Restarting Claude Code activates the patch you've already downloaded. On the memory side, finished subagent transcripts are now freed sooner instead of lingering while later subagents run, and a memory spike that happened after Claude edited a very large file is fixed.
Startup and first-command time also improve, because PATH lookups now resolve without spawning a helper process.
On to the broader Anthropic news. CNBC has a piece on the Fable five point two and Mythos five point two releases -- the same models now sitting behind your Claude Code sessions -- and frames them against Dario Amodei's public call to slow frontier development. That tension is showing up in the courts too. Reuters reports that OpenAI, Google, and Anthropic have been hit with a lawsuit asking a court to halt frontier model development at all three labs.
The outcome would bear directly on the cadence of model releases you see in Claude Code. There's financial scrutiny as well. The New York Times reports that Wall Street is examining Anthropic's disclosed compute and infrastructure commitments ahead of a potential listing. Those spending plans set the capacity that determines rate limits and model pricing for the tools you use every day.
And on the product side, The Verge reports that Anthropic is extending Claude into documents, meetings, and workplace tasks -- an office suite, essentially. The artifact and connector plumbing behind it is the same machinery showing up in Claude Code's permission prompts. One last note before we go: if you rely on the Bash sandbox or a third-party gateway, the permission flow and context window defaults are the two changes most likely to alter how your sessions behave this week. That's it for the brief.
I'm Shannon, and we'll see you next week.
Show Notes
- Code Mode documentationcode.claude.com
- Anthropic releases first frontier models since Amodei's call for slowdown as IPO buzz buildscnbc.com
- Wall Street Braces for Anthropic's Spendingnytimes.com
- Anthropic has its own AI-powered office suite nowtheverge.com
- OpenAI, Google, Anthropic hit with lawsuit seeking halt to frontier AI developmentreuters.com
