OpenCode with UberLLM: a cheaper provider in opencode.json

Steps

1. Make a key on your keys page and add credit.
2. Open (or create) ~/.config/opencode/opencode.json. For one project only, use opencode.json in the project folder instead.
3. Add the provider:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "uberllm": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "UberLLM",
      "options": {
        "baseURL": "https://api.uberllm.dev/v1",
        "apiKey": "{env:UBERLLM_API_KEY}"
      },
      "models": {
        "moonshotai/kimi-k3": { "name": "moonshotai/kimi-k3" },
        "deepseek/deepseek-v4.1-flash": { "name": "deepseek/deepseek-v4.1-flash" },
        "deepseek/deepseek-v4.1-flash": { "name": "deepseek/deepseek-v4.1-flash" }
      }
    }
  },
  "model": "uberllm/moonshotai/kimi-k3",
  "small_model": "uberllm/deepseek/deepseek-v4.1-flash"
}

4. Give OpenCode the key. Either set the variable the config reads:

export UBERLLM_API_KEY=ull_your_key

or run /connect inside OpenCode, scroll to Other, enter uberllm as the provider id and paste the key. The id must match the name in provider exactly. If you use /connect, you can delete the apiKey line.

5. Start opencode, type /models and pick one of the UberLLM models.

Why the model id has two slashes

OpenCode names a model <provider>/<model>, and UberLLM ids already contain a slash (deepseek/…). OpenCode splits on the first slash only, so uberllm/moonshotai/kimi-k3 means provider uberllm, model moonshotai/kimi-k3. The key under models must be the UberLLM id exactly as it is on uberllm.dev/models.

Tell OpenCode the context size

OpenCode uses the context size to know when to compact a long session. Add a limit to each model, using the numbers on its model page:

"moonshotai/kimi-k3": {
  "name": "moonshotai/kimi-k3",
  "limit": { "context": 1000000, "output": 64000 }
}

The numbers above are an example; copy the real ones for the model you choose.

Use the right package

Keep "npm": "@ai-sdk/openai-compatible". That package calls /v1/chat/completions, the route every UberLLM model is tested on.

Troubleshooting

• ProviderModelNotFoundError: the model you picked is not in the models list, or the key there differs from the id in "model". Add it with the exact UberLLM id.
• 401 / no answer: {env:UBERLLM_API_KEY} becomes empty when the variable is not set in the shell that started OpenCode. Export it, or use /connect.
• /connect key not used: the provider id you typed there is not uberllm. Run opencode auth list to see what is stored.
• 404: check baseURL ends in /v1, and the npm package is the openai-compatible one.
• 402 insufficient credits: add credit on the credits page.

Keep reading

• Reference page: OpenCode doc
• Kilo Code setup (same config format)
• Claude Code on UberLLM
• Cheapest Claude API