Claude Code v2.1.266: Fable 5.1 becomes the default Fable model and a fullscreen /diff panel#

Part of the Claude Code Version Tracker series. | Official Env Vars | Official Changelog

Coming from v2.1.252, a v2.1.266 user gets Claude Fable 5.1 (claude-fable-5-1) as the default Fable model with a 1M-token context window and $10/$50 per Mtok pricing ($0.25/Mtok on cache reads), and a /diff panel that opens beside the conversation in fullscreen and shows uncommitted changes as Claude edits.[1] Version 2.1.266 on its own is a single fix: the undocumented CLAUDE_CODE_USE_GATEWAY variable is ignored again when set alone, undoing a 2.1.265 regression that forced Cloud-gateway sign-in and made API-key, apiKeyHelper, and custom-auth setups fail every request with "Not signed in to the Cloud gateway."[1]

Fable 5.1 lands, then gets sanded down#

Fable 5.1 arrived in v2.1.257 as the new default Fable model.[1] The releases after it fixed the rough edges a new model exposes: the /model picker now shows Fable 5.1 for organizations that can use it instead of only accepting /model claude-fable-5-1 typed by hand, prompt caching now covers the context attached after tool results (so it is no longer re-sent uncached every tool-call turn), changing effort mid-session no longer invalidates the cache, and auto-compact for 1M models now compacts shortly before the token limit rather than timing out.[1] Claude apps gateway sessions still resolve fable and best to Fable 5 until a gateway is configured for 5.1; pick it explicitly in /model to use it there.

Working faster inside a session#

/diff toggles a panel in fullscreen mode that shows your uncommitted changes as Claude edits them.[1] /cost and the status line's prompt_cache field now name the likely cause of a cache miss (tool definitions or system prompt changed, idle past the TTL).[1]

Output handling grew room and guardrails. bashOutputMaxChars and taskOutputMaxChars raise how much command and background-task output reaches Claude inline before it spills to a file, up to 128K characters, and tool results saved to disk are now capped at 1 GB with an in-conversation note when a saved file was truncated.[1] For headless and unattended hosts, --permission-prompts none denies anything that would prompt while the active permission mode keeps deciding, --append-subagent-system-prompt-file reads an oversized subagent prompt from a file, and managedMcpServers lets an organization hand every user the same HTTP/SSE MCP servers.[1]

The auto-mode trust boundary tightened#

Several changes narrow what auto mode approves on its own. A Containment Escape rule stops cloud metadata-credential fetches, egress evasion, and cross-tenant reach from being auto-approved unless your environment marks them expected, and a one-time prompt now fires before the first file read outside the working directories, with permissions.blockReadsOutsideWorkingDirectories to block such reads entirely.[1] Reading an artifact that is not yours always asks first, even in auto mode, and when Claude summarizes an artifact someone else wrote it now treats the page as untrusted content and flags embedded instructions rather than relaying them. A link that packs content into a public diagram renderer's URL is now treated as an upload to that site and is no longer auto-approved.[1]

What these tell us#

The model surface is the clearest direction. Fable 5.1 became the default Fable model in v2.1.257, and later versions in the window fixed prompt caching for the context attached after tool results, effort-change cache invalidation, and auto-compact for 1M-context models.

The auto-mode changes move in one direction: fewer actions cleared without asking. Credential fetches from cloud metadata, reads outside the working directory, and reads of artifacts written by someone else all now stop for confirmation, and untrusted artifact text is flagged rather than passed through as instructions.

This analysis is conducted for educational and research purposes under fair use principles. All trademarks and software referenced belong to their respective owners. This content is not intended to infringe on any intellectual property rights, circumvent any protections, or encourage unauthorized access to proprietary systems.

Sources#

  1. Claude Code Official Changelog, v2.1.253 through v2.1.266 release notes

Let's Connect

BlueskyLinkedInThreads
← Back to all posts