Claude Code Received Three Updates in One Day: Model Picker, Security, and tmux Fixes

Anthropic on June 12, 2026, released three consecutive Claude Code updates within just 20 hours and 37 minutes — v2.1.174 at 01:16 UTC, v2.1.175 at 04:23 UTC, and v2.1.176 at 21:53 UTC. Each carries its own changes, but together they trace a clear direction: stricter model control in enterprise environments, more precise cost attribution, and fewer edge-case bugs disrupting everyday work.

Why Did Anthropic Release Three Versions on the Same Day?

The reason is not dramatic — this is a standard continuous-delivery cadence that is typical for a CLI tool like Claude Code. Each version closes a clearly defined set of changes, and users receive fixes quickly without waiting for a large quarterly release.

v2.1.174: Model Picker and Billing Fixes

The first version addresses several visible annoyances. The model picker (/model) now correctly displays Opus as a distinct row on Max, Team Premium, and Enterprise plans, and Sonnet on Pro and Team plans — previously the model family that the Default resolved to could be hidden. The picker also showed the wrong label when ANTHROPIC_DEFAULT_SONNET_MODEL pointed to a different Sonnet variant than the default.

Enterprise administrators will particularly welcome one fix: the banner announcing “Fable 5 now uses billable credits” was incorrectly shown to users on usage-based billing plans for which this was not true. It now appears only when relevant.

For development teams using Amazon Bedrock in government regions (us-gov-*), the model invocation prefix has been corrected — instead of the erroneous global prefix, the correct us-gov is now used, eliminating 400 errors in API calls.

The version also introduces a practical new setting: wheelScrollAccelerationEnabled disables scroll acceleration in fullscreen mode. VS Code users gain a more detailed view in the Account & usage dialog — cache miss, long context, and a breakdown by skills, agents, plugins, and MCP servers are now visible for the past 24 hours or 7 days. A 1–2 second pause on exit from Claude Code after a shell command was interrupted has also been eliminated.

v2.1.175: enforceAvailableModels

A brief but security-significant release introducing one new managed setting. enforceAvailableModels allows IT administrators to make the availableModels list a true hard boundary — a Default model that would resolve to a disallowed model automatically falls back to the first permitted one. User or project settings cannot extend the list beyond what the administrator specifies. This closes the door to scenarios in which a user could bypass organizational model policies through local configuration files.

v2.1.176: Languages, tmux, and Bedrock

The third version is the most comprehensive. Claude Code session titles are now automatically generated in the language of the conversation, and a new language setting allows pinning to a specific language regardless of content.

A bug that had troubled users working in tmux inside an SSH session is now fixed — /copy and mouse-based copying now reach the system clipboard. Bedrock users gain smarter credential caching: instead of a fixed one-hour expiry, credentials from awsCredentialExport are now cached until the token’s actual expiration time. A fix for availableModels closes another loophole — model aliases can no longer be redirected to a blocked model via ANTHROPIC_DEFAULT_*_MODEL environment variables.

Additional fixes include: correct behavior for the back key in claude agents, a fix for the /cd command that left stale git branch data after changing directories, a fix for an authentication error in cloud sessions, and improved handling of network paths on Windows.

The three versions together demonstrate that Anthropic actively follows community feedback and does not wait long before addressing issues that affect everyday work. For teams managing Claude Code deployments within organizations, enforceAvailableModels combined with the existing availableModels lists now provides complete control over which model employees can use — with no loopholes to bypass it.