Lifestyle
Claude Code 2.1.321 Update: Fable 5.2 Becomes the Default Model, With Changes to mods Permissions and Enterprise Management Rules
Anthropic's official Claude Code changelog says version 2.1.321 adds Claude Fable 5.2 and makes it the default Fable model. This release also changes mods permissions, Remote Control, managed marketplaces and gateway behavior. Both general users and enterprise administrators may be affected.
About 5 min read
New Model: Claude Fable 5.2
According to the official Claude Code changelog, version 2.1.321 adds Claude Fable 5.2 (claude-fable-5-2) and makes it the default version of the Fable model. The changelog says it offers a 1M context window and is priced at $8/$40 per million tokens, with cache reads at $0.20 per million tokens. Context is the amount of content a model can consider at once. A token is the unit a model uses to measure text. A cache read means reusing content that has already been processed.
Impact on existing users: according to the changelog, a saved Fable 5.1 model selection is changed to fable. This takes effect immediately on Pro and Max plans, and on Team and Enterprise plans once the organization allows it. To go back to the older version, the changelog says you can switch with /model claude-fable-5-1.
Permissions and Security: mods Behavior Has Changed
mods are components that Claude can write and load itself. The changelog says that in auto and bypass modes, mods written by Claude are enabled without asking. These two modes reduce or skip per-action confirmation. This is the case even when managed settings specify allowUnsandboxedCommands: false or allowManagedDomainsOnly. In dontAsk mode, these mods are not loaded.
In addition, according to the changelog, mods' own tool calls ($.tool.call, $.mcp.call) no longer trigger permission prompts. This holds even under user-defined ask rules, dontAsk, or -p runs. Deny rules and organization policy still apply. For general users, this means confirmation prompts may become less frequent. Teams with control requirements should check the changelog to confirm these changes comply with their rules.
Changes Enterprises and Administrators Should Note
- According to the changelog, managed extraKnownMarketplaces (the list of plugin sources designated by an organization) has three changes: a declared name cannot be added from another source, entries with the same name from elsewhere are hidden, and plugins listed by a declared marketplace are reinstalled from it.
- According to the changelog, the Bedrock, Vertex and Foundry upstreams of the Claude apps gateway ignore ANTHROPIC_*_BASE_URL variables in the gateway environment. To use a different endpoint, set base_url on the upstream instead.
- According to the changelog, under policy_mode: layered, the managed MCP servers from every matching policy now add up. An MCP server is a service that provides tools to Claude. Gateways already using this mode may offer more servers after upgrading.
- According to the changelog, Remote Control sessions stay in the folder they were started in, and connected clients can no longer move a session to another folder.
Other Changes at a Glance
| Item | Change described in the changelog | Who is affected |
|---|---|---|
| claude remote-control | Adds a --settings file parameter applied to every session it starts; after the file changes, a restart is needed before new sessions start | Remote Control users |
| VS Code | When both an auth token and an API key are set, a warning is shown above the prompt box | VS Code users |
| Plan names | Related messages and settings descriptions now list Ultra in addition to Pro and Max | Paid plan users |
| Headless (no interactive interface) sessions | mods reload when their files change; set CLAUDE_CODE_PLUGIN_DIR_WATCH=0 to turn this off | Automation or headless users |
| Background command notifications | Commands over 1,000 characters show only the beginning and end, to save context | All users |
| Windows | When locating programs, skips relative and UNC PATH entries and links pointing to network shares, and reads PATH changes in settings | Windows users |
| Desktop Code tab | Number of messages that can be sent to other sessions raised from 10 to 20 | Claude Desktop users |
FAQ
What is the main change in 2.1.321?
According to the changelog, this release adds Claude Fable 5.2 (claude-fable-5-2) and makes it the default Fable model, with a 1M context window.
How much does Fable 5.2 cost?
The changelog lists a price of $8/$40 per million tokens, with cache reads at $0.20 per million tokens.
What happens to people who had selected Fable 5.1?
According to the changelog, the saved selection is changed to fable. This takes effect immediately on Pro and Max plans, and on Team and Enterprise plans once the organization allows it. You can switch back with /model claude-fable-5-1.
Do mods now show fewer permission prompts?
The changelog says mods' own tool calls no longer trigger permission prompts, but deny rules and organization policy still apply. Mods written by Claude are not loaded in dontAsk mode.
What should enterprise administrators watch for before upgrading?
According to the changelog, the source rules for managed extraKnownMarketplaces are stricter. The gateway's Bedrock, Vertex and Foundry upstreams must set endpoints through base_url. Under policy_mode: layered, managed MCP servers from every matching policy add up, so gateways already using that mode may offer more servers.
Browse the latest news in this topic
Lifestyle
Nvidia Board Investigates CEO Jensen Huang Over Ties to Megaspeed Executive Alleged to Be a Chinese Spy
The Verge reports that Nvidia has announced an investigation into CEO Jensen Huang's relationship with Megaspeed executive Alice Huang, whom US officials call a Chinese spy. The company says it sees no reason to believe national security or its secrets were compromised; the probe continues.
Lifestyle
Anthropic Publishes a Claude Model Guide: How Haiku, Sonnet, Opus and Fable Divide the Work
A post on the claude.dev blog explains what each of Claude's four general-purpose models is suited for, how much their API prices differ, and what to check before switching. For anyone building products or writing code with Claude, the choice affects cost, speed and results. Everything below is as stated by that official source.
Lifestyle
AWS Adds NVIDIA Nemotron 3.5 Super VL: SageMaker JumpStart Users Can Deploy an AI Model That Reads Images and Video
AWS announced that NVIDIA's Nemotron 3.5 Super VL, which understands images, video and text, is now available in Amazon SageMaker JumpStart in NVFP4 and BF16 versions, for processing documents, charts, video and screen content. This article covers what it does, how the two versions differ, performance figures and limits.
Lifestyle
Amazon Bedrock Managed Knowledge Base Adds Single-Shot RAG and Retrieval-Only Modes, Letting Developers Match Retrieval to Question Difficulty
AWS announced that Amazon Bedrock Managed Knowledge Base now supports single-shot RAG and retrieval-only requests through AgenticRetrieveStream, alongside its existing agentic retrieval. Developers building internal company AI assistants can handle quick lookups and complex comparisons with one API. AWS says simple questions can get faster responses with less model usage.
Sources
- 2.1.321 · Checked: