Using Anthropic's models via Vertex AI in opencode

major@mhtx.net (Major Hayden) — Mon, 13 Apr 2026 00:00:00 +0000

We use Vertex AI at work for accessing Anthropic’s models and the opencode setup is quite smooth! Although getting your authentication credentials from Google can be a little tricky, using them with opencode involves setting a few environment variables.

Setting up Vertex AI

Before you point opencode at Vertex AI, you need a few things on the GCP side.

Prerequisites

A GCP project with the Vertex AI API enabled
IAM permissions for the Claude models (your account needs the aiplatform.endpoints.predict permission, or a broader role like Vertex AI User)
Either a service account key or Application Default Credentials (ADC)

Authentication

You have two main options here. The easiest for local development is ADC:

gcloud auth application-default login

This stores credentials locally and opencode picks them up automatically.

If you’re using a service account (common in enterprise setups where your org provisions access), point to the key file:

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/your-service-account.json

Environment variables

Set these in your shell profile or session:

export ANTHROPIC_VERTEX_PROJECT_ID=your-gcp-project-id
export GOOGLE_CLOUD_PROJECT=your-gcp-project-id
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/your-service-account.json # if not using ADC

You can also set VERTEX_LOCATION if you need a specific region, but the global endpoint is the default. It routes to the nearest available region.

Using opencode

You don’t need to configure opencode at all! It finds the environment variables and takes care of the rest for you.

You can type /models, press enter, and you should see the list of available models from Vertex AI. From there, use CTRL-F to select any favorites that you use often.

Troubleshooting

If you don’t see any models, double check the environment variables to ensure they are set in your terminal session. You may need to open a new terminal or source your shell profile to pick up the new variables.

You can also run opencode models --refresh to force a refresh of the model list from Vertex AI.