Keep three values together.
Use the Codimixa base URL, a Codimixa API key and an available model ID in the same configuration.
| Setting | Value |
|---|---|
| baseURL | https://api.codimixa.com/v1 |
| apiKey | Your Codimixa key, read from your local environment |
| Model ID | Copy the exact ID from the Codimixa model page |
Choose the config file.
Put opencode.json in your project root, or use ~/.config/opencode/opencode.json for a user-wide default. Matching project settings override global settings. OPENCODE_CONFIG selects a custom file; check project and managed settings if your choice is overridden. These instructions follow the OpenCode documentation checked October 25, 2026.
Add the connection to OpenCode.
Use @ai-sdk/openai for GPT models that use Responses, including GPT-6 Astra and GPT-6.1 Sol. Save this example in opencode.json and set CODIMIXA_API_KEY in your local environment.
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"codimixa": {
"npm": "@ai-sdk/openai",
"name": "Codimixa",
"options": {
"baseURL": "https://api.codimixa.com/v1",
"apiKey": "{env:CODIMIXA_API_KEY}"
},
"models": {
"gpt-6-astra": {
"name": "GPT-6 Astra"
}
}
}
},
"model": "codimixa/gpt-6-astra"
}Choose the adapter for another model.
Use @ai-sdk/openai-compatible for Chat Completions models such as deepseek-v4.1-flash. Change both the models key and the final model value to the exact Codimixa ID. GPT-6.1 Sol requires Responses for tool calling.
Set the default model or switch for one run.
The model value is provider_id/model_id: codimixa/gpt-6-astra in this example. Both keys must exist in your provider configuration. Use /models to switch interactively; an explicit --model option takes priority over the configured default.
opencode --version
opencode --model codimixa/gpt-6-astraTry text first, then an editing task.
Ask OpenCode to explain one function. Next, ask for a small edit with a test you can run locally. Review the patch and inspect your account usage. Keep the client version and exact model ID with the result before using the setup for a larger change.
Resolve the common setup mistakes.
Start with the error from one small request. Repeating a configuration error adds no useful evidence.
| Symptom | Check next |
|---|---|
| Authentication fails | The environment variable is available to the OpenCode process and contains a Codimixa key. |
| Model is missing | The provider key, models key and provider/model value agree exactly. |
| Wrong default starts | Check --model, project config and managed settings before changing the global file. |
| Text works, tools fail | Match the model’s recommended API format to the adapter and inspect the response. |
| Unexpected usage | Inspect all turns and billed categories for the task, including retries. |
Keep the price visible as context grows.
Selected model rates start at 0.3× the matching official API reference. Longer sessions still send more context; compare the model’s actual rates with a representative coding session.
A FEW DETAILS
Before you continue
Should I put /chat/completions in the base URL?
For this configuration, use https://api.codimixa.com/v1 as the base URL. The selected adapter supplies the request path.
Which adapter sends Responses requests?
Use @ai-sdk/openai for /v1/responses. Use @ai-sdk/openai-compatible for /v1/chat/completions. The example above selects Responses for GPT-6 Astra.