Claude Code Data Residency Workspaces Now Carry a 1.1× Cost Premium Operators Must Route Around

Two features landed quietly in Claude Code 2.1.239 that have real billing and governance consequences for teams running managed deployments: cost estimates now include the 1.1× US-only-inference premium for data-residency workspaces, and a new /claude-api upgrade command handles the Python SDK migration from anthropic 0.x to 1.x — including a breaking change in how timeouts are constructed. A third change, plugin sync visibility, gives operators a clearer picture of which connectors are managed by their org. Claude Code 2.1.245 then shipped a narrowly targeted fix for Linux distributions shipping glibc 2.44 that crashes Claude Code on startup.

Taken together, the billing transparency change is the one most likely to trigger a routing policy review.

What the 1.1× data-residency premium actually changes for operators

Anthropic's data-residency workspaces — teams that elect US-only inference to satisfy compliance or data-handling requirements — pay a 1.1× multiplier on all token pricing. That has been documented on the API pricing page for a while. What changed in 2.1.239 is that Claude Code now reads the workspace configuration and folds the premium into every cost estimate: /cost output, the per-turn status line, and telemetry cost figures all reflect the actual contracted rate rather than the list price.

The operational consequence is predictable: teams that previously eyeballed Claude Code's cost output against list-price benchmarks will now see numbers that are 10% higher. Budget alerts may fire earlier. Cost-per-task comparisons against Bedrock-routed or Vertex-routed deployments will show a gap that was always there but invisible in the tool.

For routing policy, this matters because data-residency is a workspace-level switch, not a per-request flag. A team that turns on US-only inference applies the multiplier to every request flowing through first-party Claude Code. Teams routing the same models through AWS Bedrock or Google Vertex AI do not pay the Anthropic data-residency surcharge — they may pay cloud-specific fees instead, but those are controlled separately.

This creates a meaningful cost-routing decision point: if compliance requirements are satisfied by Bedrock's or Vertex's own data-handling guarantees, routing through those providers avoids the first-party premium. The tradeoff is that provider-level features — managed settings, org-managed plugins, console analytics — work only on the first-party path.

The Python SDK upgrade path: what breaks and what the command fixes

Claude Code 2.1.239 also shipped /claude-api upgrade for Python projects. The command migrates anthropic 0.x usage to 1.x. The specific breaking change called out in the changelog is timeout construction: httpx.Timeout no longer works in 1.x — the correct form is anthropic.Timeout.

# anthropic 0.x — no longer valid in 1.x
import httpx
client = anthropic.Anthropic(timeout=httpx.Timeout(60.0, connect=5.0))

# anthropic 1.x — correct form
client = anthropic.Anthropic(timeout=anthropic.Timeout(60.0, connect=5.0))

The migration matters for teams running Claude Code through a gateway with custom timeout policies. If you proxy Claude Code API calls through a request layer that sets httpx.Timeout objects, those will fail silently on 1.x until you update the call site. The /claude-api upgrade command scans the project and rewrites the relevant patterns — run it in any project where you've pinned anthropic 0.x and are evaluating an upgrade.

Plugin sync visibility for operator governance

In the same release, plugins synced from claude.ai now appear in /plugins with a @synced suffix — for example linear@synced. The enable/disable commands accept the @synced qualifier so operators can toggle synced connectors explicitly without conflating them with locally installed plugins of the same name. Critically, synced plugins never override a locally installed plugin of the same name, which removes a class of accidental override bugs in managed deployments.

For operators managing Claude Code at scale through a managed-settings layer, this gives a clearer audit surface: /plugins output now distinguishes org-managed connectors from user-installed ones, and the managed marker shows in both /mcp and /plugins for connectors whose authentication is handled by the org.

What 2.1.245 fixes for self-hosted runner deployments

Claude Code 2.1.245 is a targeted fix for a startup crash on Linux distributions shipping glibc 2.44 — specifically Arch Linux, CachyOS, and Fedora Rawhide. The crash was a native binary compatibility issue, not a logic error. If your Claude Code runners are on Debian, Ubuntu LTS, or RHEL-family distributions, you are unlikely to have been affected. Teams running bleeding-edge Arch or Fedora Rawhide builds as CI runner images should update.

The routing decision this batch exposes

The 1.1× cost premium is now visible in tooling. Teams can measure it. The provider routing question that follows is: does your compliance posture actually require first-party US-only inference, or would Bedrock's or Vertex's own data handling cover the requirement?

If Bedrock or Vertex satisfies your compliance bar, routing through those providers removes the first-party surcharge and keeps your Claude Code cost estimates aligned with Bedrock/Vertex pricing — no 1.1× markup. The Claude Code 2.1.239 Bedrock billing fix from last week is relevant context here: teams that moved to Bedrock to avoid the data-residency premium should verify they are not running behind a proxy that strips response Content-Type headers, which was triggering the double-billing bug.

For teams already on first-party Claude with data-residency enabled, the surfaced cost estimate at least makes the premium legible. You can now set modelPricing in managed settings (added in 2.1.243) to reflect your contracted rate, so cost attribution across teams remains accurate even after the multiplier is applied.

The Claude Code 2.1.243 article covers the managed settings changes — modelPicker, promptCacheTtl, and modelPricing — that combine with this billing-transparency change to give operators a more complete cost-control layer across a managed Claude Code deployment.