Claude Code v2.1.243: Loops Breakdown in /usage, a Curated /model Picker, and Prompt Cache TTL Settings

Claude Code v2.1.243 is a large release with 60 changelog entries. It adds a Loops breakdown to /usage, new settings for curating the /model picker, setting prompt cache TTLs, and applying an organization’s contracted model pricing, plus a keyless Console sign-in under /login. It also shrinks the native download, lowers memory use, and improves startup time, and fixes issues across MCP, auto mode, cross-session messaging, Remote Control, and the VS Code extension.

What’s New

Loops Breakdown in `/usage`

/usage now includes a Loops breakdown with each loop’s run count, total tokens, tokens per run, and last run, so runaway or chatty /loop tasks are easy to spot. In the terminal, consecutive /loop wake-ups where Claude has nothing to do now fold into a single line instead of printing each one.

Curate the `/model` Picker With `modelPicker`

The new modelPicker setting curates the /model picker with an ordered, labeled list of models. Entries can use any id spelling, including Vertex and Bedrock ids, and the list is either appended to the built-in lineup or replaces it.

Prompt Cache TTL Settings

The new promptCacheTtl and subagentPromptCacheTtl settings let API-key and cloud-provider users keep a 1-hour prompt cache on the main conversation while subagents stay at 5 minutes.

Contracted Pricing With `modelPricing`

The new modelPricing managed setting makes /cost, the status line, and telemetry cost figures use an organization’s contracted per-model rates and discount multiplier instead of list price.

Under /login → Anthropic Console, a new “Sign in with your Console account” option (recommended) sits alongside creating an API key, so organizations that don’t allow API keys can sign in. /login over SSH is also improved: the sign-in URL appears immediately, pressing c reports how the URL was copied instead of always claiming success, and a hint explains how to select text in fullscreen.

Managed Settings Visibility

/status has a new Skipped sources line that lists managed settings sources (for example managed-settings.json) that are present but not applied because a higher-precedence managed source is active. In /mcp and /plugins, claude.ai connectors whose authentication is managed by your organization now show a managed marker.

GitHub Status for Claude Code on the Web

A new /status line shows whether GitHub is connected for Claude Code on the web (Pro/Max) and points to /web-setup when it isn’t, and claude.ai users who haven’t connected GitHub for Claude Code on the web now get a tip pointing to /web-setup. /web-setup also no longer repeatedly asks you to log in when an older GitHub CLI (without gh auth token) is already authenticated.

Subagent Model and Effort in `/tasks`

/tasks and the agent detail dialogs now show the model (and effort level) each subagent ran on.

Smaller Download, Lower Memory, Faster Startup

The native install and auto-update download is now a zstd-compressed binary of about 75 MB instead of 340 MB on Linux x64, and the binary is about 2 MB smaller because the bundled skill and prompt text is stored more compactly. Native builds now load code on demand instead of keeping the whole bundle resident (roughly 40–70 MB less memory per session), and in long-running sessions the runtime garbage-collects sooner as the heap grows, improving peak memory usage. Startup time is also improved: sandbox and MCP bring-up no longer block the first frame, bare launches skip subcommand registration, and workflow discovery, settings, and trust-store work is cheaper.

Sandboxed Bash No Longer Lists Allowed Hosts

The sandboxed Bash tool prompt no longer lists allowed network hosts, so Claude attempts requests (and you can approve new hosts) instead of assuming unlisted hosts are blocked.

`/model`, `/fast`, and `/effort` Run Immediately on More Setups

On Bedrock, Vertex, and Foundry, and when telemetry is disabled, /model, /fast, and /effort now also run immediately instead of queueing until the turn ends.

Computer Use on macOS Requires Granting Finder

With computer use on macOS, clicking the desktop, Dock, or a Finder window now requires granting Finder via the access dialog, like any other app.

Fixes and Improvements

MCP: Remote MCP servers in non-interactive (-p) and SDK sessions no longer fail to recover after a dropped connection; they now reconnect automatically or report as failed. MCP server sign-in started from the desktop app no longer fails with “Invalid redirect URI” on servers that support client ID metadata documents, such as Linear.
Auto mode: Auto mode no longer stays unavailable at startup when a temporary server-side disable was cached and later flag fetches failed. Auto mode tool calls are no longer denied as “temporarily unavailable” after about a minute of waiting when the API was briefly overloaded and asked the client to retry.
Unresponsive API: Sessions no longer go silent for 10+ minutes when the Anthropic API never starts a response; the request now times out after about 3 minutes, retries once, then shows API Error: No response from API.
/model picker: Picking Ultracode in the /model picker now applies it to the current session instead of being silently ignored. The picker and the bundled claude-api skill now show Sonnet 5’s $2/$10 per Mtok pricing as its standard list price rather than a limited-time promo.
Resume: /resume no longer lists only the 50 most recent sessions; the picker now loads more as you scroll.
Background subagents: Background subagents now wake when their last background Bash task completes.
Cross-session messaging: Cross-session messaging no longer silently turns off inside user namespaces and rootless containers after the 2.1.232 socket-directory hardening. The inbox socket now closes connections that send no complete line within 30 seconds.
Remote Control: claude remote-control no longer exits and strands attached Remote Control sessions when the server drops its environment mid-session; it now recovers.
Hooks: Hook if conditions like Bash(cat *) no longer fire on unrelated Bash commands when the command contains $() or backtick command substitution followed by more arguments.
Agent definitions: --agents no longer silently ignores invalid JSON or invalid agent definitions; it now exits with a clear error, like --mcp-config.
CI authentication: With workload identity federation in CI, processes in one job now share the exchanged token instead of re-exchanging the single-use token, and a rejected exchange fails fast with the server’s message.
Sandbox: Sandbox network-violation details are no longer dropped from the Bash tool result when the blocked command still exits 0, for example when curl prints the proxy’s 403 page.
Rate limits: The status line rate_limits fields and /usage no longer keep showing a rate-limit window’s pre-reset usage percentage after the window resets while the session is idle.
Input history: Ctrl+R history search and up-arrow history no longer break when ~/.claude/history.jsonl contains a malformed entry.
Claude in Chrome: Claude in Chrome no longer loses its connection to Claude Code after an auto-update cleans up the version it was set up with; the native host now launches via the stable claude launcher.
VS Code: Sessions started before feature flags were first fetched (for example right after install) now open in auto mode or your configured default mode instead of the default permission mode.

This release contains 60 changes in total; see the official changelog linked below for the complete list.

How to Update

Check your installed version: claude --version
Update in place: claude update
If you have scripts that post to the cross-session messaging inbox socket, make them connect only once their data is ready, because the socket now closes connections that send no complete line within 30 seconds.

What’s New

Loops Breakdown in /usage

Curate the /model Picker With modelPicker