OpenCode config for Claude: opencode.json setup, models and common errors

OpenCode reads its settings from opencode.json. To run it on Claude through this gateway you change one value, the Anthropic provider's baseURL, and give it a key. Below are two working configs, the model and context settings that matter, and fixes for the errors OpenCode users run into on this gateway, taken from our own request logs.

Where opencode.json lives

OpenCode merges a global config with a per-project one: ~/.config/opencode/opencode.json for settings you want everywhere, and opencode.json in a project root for that repository. Adding the $schema line gives you autocompletion and validation in most editors.

Keep the key out of the file. Put it in an environment variable and reference it as {env:AIPRIME_API_KEY}. Keys you add with /connect inside OpenCode are stored in ~/.local/share/opencode/auth.json; opencode auth list shows what is stored there.

export AIPRIME_API_KEY=your_gateway_key   # add to ~/.bashrc or ~/.zshrc
opencode auth list                         # credentials OpenCode already knows about

Option 1: point the built-in Anthropic provider at the gateway

The smallest change: keep OpenCode's own Anthropic provider and its model list, and override where it sends requests.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "anthropic": {
      "options": {
        "baseURL": "https://aiprimetech.io/v1",
        "apiKey": "{env:AIPRIME_API_KEY}"
      }
    }
  },
  "model": "anthropic/claude-sonnet-5",
  "small_model": "anthropic/claude-haiku-4-5"
}

Option 2: a custom provider with your own model list

A separate provider entry keeps the gateway apart from any Anthropic login, lets you list exactly the models you want in the picker, and lets you set context limits per model.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "aiprime": {
      "npm": "@ai-sdk/anthropic",
      "name": "AI Prime Tech",
      "options": {
        "baseURL": "https://aiprimetech.io/v1",
        "apiKey": "{env:AIPRIME_API_KEY}"
      },
      "models": {
        "claude-sonnet-5":  { "name": "Claude Sonnet 5",  "limit": { "context": 200000, "output": 64000 } },
        "claude-opus-5-5":  { "name": "Claude Opus 5.5",  "limit": { "context": 200000, "output": 64000 } },
        "claude-haiku-4-5": { "name": "Claude Haiku 4.5", "limit": { "context": 200000, "output": 32000 } }
      }
    }
  },
  "model": "aiprime/claude-sonnet-5",
  "small_model": "aiprime/claude-haiku-4-5"
}

Prefer @ai-sdk/anthropic for Claude: it speaks the native Messages API, so prompt caching and thinking work as Anthropic designed them. The gateway also serves the OpenAI shape, so @ai-sdk/openai-compatible with the same https://aiprimetech.io/v1 base URL works too, and it is the route to use for GPT models on the same key.

Models and context limits

model is what the main agent uses; small_model handles cheap side jobs such as session titles. The limit.context number tells OpenCode when to compact the session. If a long session fails with prompt is too long, lower it (200000 is a safe value) so OpenCode summarises earlier instead of sending more than the endpoint accepts.

Model IDs are exact strings. Older names such as claude-3-5-sonnet-20241022 or dated snapshots are not served; use the IDs above or the list in your dashboard.

Check the key and URL before starting OpenCode

One request from the terminal tells you whether the key and URL are right, independently of OpenCode's config:

curl -s https://aiprimetech.io/v1/messages \
  -H "x-api-key: $AIPRIME_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-haiku-4-5","max_tokens":20,"messages":[{"role":"user","content":"ping"}]}'

A JSON reply with a content array means the gateway side is fine and any remaining problem is in opencode.json. Invalid API key means the variable is empty or holds the wrong key.

Troubleshooting: errors OpenCode users hit

These are the errors that show up most for OpenCode on this gateway, with the fix for each.

For errors that come from the model side rather than the client, see the Claude API errors guide.

Frequently asked questions

AI Prime Tech is an independent API gateway. It is not affiliated with, endorsed by, or a reseller of Anthropic. Claude and related model names are trademarks of their respective owners.