Before migrating to the Copilot SDK, my cli tools used the Copilot CLI directly.
When using models like gpt-5.2-codex via the CLI:
- Copilot returned an interactive response asking the user to enable the model
- This enablement flow was handled correctly in the CLI integration
After migrating to the Copilot SDK, the same model now results in:
null response or
- request timing out
- no error
- no enablement prompt or metadata indicating the model requires activation
This makes the behavior ambiguous and impossible to handle programmatically.