Repository navigation
Agent temperature is dropped for config-defined custom OpenAI-compatible models #34554
Closed as not planned
Description
Description
When a model is defined only in opencode.json under a custom provider, agent.<name>.temperature can be silently dropped even though the request goes through the OpenAI-compatible path.
This happens because the custom-model merge path defaults capabilities.temperature to false when the config entry omits that field. Session request construction then strips temperature before the provider request is built.
Steps to reproduce
Use a config like:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"myprovider": {
"npm": "@ai-sdk/openai-compatible",
"name": "maas",
"options": {
"baseURL": "http://localhost:8090/v1",
"apiKey": "test-key"
},
"models": {
"MiniMax-M2.7": {
"tool_call": true,
"limit": { "context": 32000, "output": 8000 }
}
}
}
},
"agent": {
"build": { "temperature": 1 }
}
}Send a request with the build agent and inspect the outgoing OpenAI-compatible payload.
Expected behavior
The request should include the configured temperature.
Actual behavior
temperature is omitted unless the model config explicitly sets "temperature": true.
OpenCode version
1.17.11
Operating System
Windows 11
Terminal
Powershell
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Jun 30, 2026
This issue might be a duplicate of existing issues. Please check:
temperatureparameter not passed to LLM API requests for custom OpenAI-compatible providers, identical config structure (myprovider/MiniMax-M2.7,@ai-sdk/openai-compatible), same version and OS.SeashoreShi commentedon Aug 8, 2026
Hi, I've opened a fix for this: #41016. It defaults the temperature capability to enabled for config-defined custom models (explicit
temperature: falsestill wins), and includes regression tests verifying the agent temperature is forwarded in the request body. Verified locally: 129 pass / 0 fail across the provider and llm suites.github-actions commentedon Oct 7, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.