Description
The model ID used in a session auto-switches silently (no error, no prompt) to a different model, both at restart time and mid-conversation. There is no fixed pattern — switches observed between OpenAI models and DeepSeek models, and vice versa.
Steps to Reproduce (observed, not perfectly consistent)
- Start a session with an explicitly set model.
- Either restart OpenCode or continue a long conversation.
- At some point, the session silently starts using a different model (e.g.
deepseek-ai/DeepSeek-V4-Flash or another model) without any notification.
Expected Behavior
The model should remain stable throughout the session unless the user explicitly changes it (via /model or config edit). Any automatic fallback or switch should be clearly communicated to the user.
Actual Behavior
The model switches silently. The user only notices because response behavior changes (e.g. sudden loss of reasoning, different response style).
Environment
- OpenCode version: latest
- Observed with
gpt-5.4-mini → deepseek-ai/DeepSeek-V4-Flash and other combinations
- No error messages in logs
Description
The model ID used in a session auto-switches silently (no error, no prompt) to a different model, both at restart time and mid-conversation. There is no fixed pattern — switches observed between OpenAI models and DeepSeek models, and vice versa.
Steps to Reproduce (observed, not perfectly consistent)
deepseek-ai/DeepSeek-V4-Flashor another model) without any notification.Expected Behavior
The model should remain stable throughout the session unless the user explicitly changes it (via
/modelor config edit). Any automatic fallback or switch should be clearly communicated to the user.Actual Behavior
The model switches silently. The user only notices because response behavior changes (e.g. sudden loss of reasoning, different response style).
Environment
gpt-5.4-mini→deepseek-ai/DeepSeek-V4-Flashand other combinations