after system upgrade, Claude Opus 4.1 is disconnected

Question

I upgraded my OS and some extensions recently, and now my V Code AI is broken. I had selected Claude Opus 4.1 for my LLM, but now it says disconnected, and then spins forever trying to get it installed - obviously not doing anything but spinning.

I've tried backing out of the latest Microsoft Foundry to the pre-release version. No dice.

Tired updating VS Code again. Restarting.

Tried choosing different models.

Broken

Answer 1

Hi Kameron Cole,

This issue usually happens because Claude Opus 4.1 is a preview model and depends on the Foundry deployment and local client setup staying in sync after a system or extension upgrade. It is not related to a service outage.

Here are a few things you can check step by step.

First, check the Claude Opus 4.1 deployment in Azure AI Foundry. Open the Azure AI Foundry portal and go to the project where Claude Opus 4.1 is deployed. Make sure the deployment exists and its state shows Succeeded. If the deployment looks stuck or failed after the upgrade, removing and recreating the same deployment often clears the disconnected state.

Second, confirm your region and subscription are still valid. Claude models in Microsoft Foundry are preview models and are only supported in specific regions like East US 2 and Sweden Central. If the resource was recreated, moved, or reconfigured during the upgrade and ended up in a different region, the model can appear as disconnected even though it is selected. Also ensure the subscription is a paid subscription, as preview partner models do not work on free or credit-only subscriptions. https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-claude

Third, verify the local tool or editor configuration after the upgrade. After an OS or VS Code upgrade, environment variables and extensions may not load correctly. If you are using Claude through VS Code or a CLI integration connected to Foundry, recheck that the Foundry resource name and model deployment name match exactly what is configured locally. A mismatch here can cause the client to spin, or show disconnected even though the deployment is healthy.

Fourth, be aware of preview model behavior. Claude Opus 4.1 is a preview model, and preview models can be more sensitive to environment changes. In some cases, switching temporarily to another Claude model like Sonnet and then switching back to Opus after verifying the deployment helps reestablish the connection. This does not affect your data or project setup.

In most cases, once the deployment state, region, and local configuration are aligned again, the disconnected issue goes away without any changes to the service itself.

Hope this helps. Let me know if you want to walk through any of the checks in more detail.

Thankyou!

Answer 2

Claude Opus 4.1 in Microsoft Foundry is a preview Anthropic model and depends on a healthy Foundry deployment and correct client configuration. After an OS and extension upgrade, the most likely causes are:

1. The Foundry deployment is not ready or is in a bad state.
2. The VS Code / Claude Code integration is misconfigured after the upgrade.

Use the following checks and fixes.

1. Verify the Claude Opus 4.1 deployment in Foundry

• In the Azure AI Foundry portal, open the Foundry resource where Claude Opus 4.1 was deployed.
• Confirm that the Claude Opus 4.1 deployment exists and its provisioning state is Succeeded.
• If the deployment shows errors or has been stuck for a long time, delete that deployment and recreate it. A similar issue has been resolved by recreating the deployment after it was stuck in a bad state.

2. Confirm subscription and region are still valid

• Ensure the subscription still meets the requirements for Claude models in Foundry (paid subscription, supported region, etc.).
• Claude models in Foundry are only available in East US 2 and Sweden Central. If the resource or deployment was moved or recreated in another region, the model will not be available and the client can appear “disconnected”.

3. Validate Claude Code / VS Code configuration If using Claude Code (CLI or VS Code extension) against Microsoft Foundry:

• Open a terminal in the project folder and run:

    claude
  

• At the Claude prompt, run:

    /status
  

• Confirm in the status output:
  • API provider: Microsoft Foundry
  • Microsoft Foundry Resource: <your-resource-name>
  • Model: Default (claude-opus-4-1) or the specific deployed model name.

If the model is missing or wrong:

• Ensure environment variables are set correctly before launching VS Code:
  • ANTHROPIC_FOUNDRY_RESOURCE must match the Foundry resource name.
  • ANTHROPIC_DEFAULT_OPUS_MODEL must match the Claude Opus 4.1 deployment name.
• If VS Code is running, close it, set the environment variables, then start VS Code from that terminal (code .) so the extension picks them up.

4. Fix common connection and auth issues Check for these common problems and apply the matching fix:

• Authorization failed (HTTP 401/403) or “Failed to get token” / “ChainedTokenCredential authentication failed”:
  • Run az login again after the OS upgrade to refresh credentials.
  • If using API key auth, ensure ANTHROPIC_FOUNDRY_API_KEY is set and valid.
• "The model <model-name> is not available on your foundry deployment":
  • Confirm Claude Opus 4.1 is actually deployed in the Foundry resource.
  • If not, deploy it again and update ANTHROPIC_DEFAULT_OPUS_MODEL to the correct deployment name.
• VS Code extension not connecting after OS upgrade:
  • Ensure all required environment variables are set in the new OS session.
  • Launch VS Code from a terminal where those variables are defined.
  • If using WSL, set the variables both on Windows and inside WSL, then from WSL run code ..
• Region errors or “model not available” after upgrade:
  • Confirm the Foundry resource is in East US 2 or Sweden Central and that Claude Opus 4.1 is deployed there.

5. If the deployment is stuck or keeps spinning If the portal shows the deployment as not ready or the client spins indefinitely even though configuration looks correct:

• Delete the Claude Opus 4.1 deployment in the Foundry portal.
• Recreate the deployment for Claude Opus 4.1 in the same supported region.
• Update the client configuration (environment variables and/or VS Code settings) to point to the recreated deployment name.

After these steps, re-run claude → /status and then send a simple test prompt to confirm the model responds.

---

References:

• Configure Claude Code for Microsoft Foundry
• Deploy and use Claude models in Microsoft Foundry (preview)
• Deployed Claude and getting API deployment resource is not ready even though it has been an hour - Microsoft Q&A