fix: update default agent init model to gpt-5.4-mini#9290
Conversation
📋 Prioritization NoteThanks for the contribution! The linked issue isn't in the current milestone yet. |
|
Azure Pipelines: Successfully started running 1 pipeline(s). 21 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run. |
There was a problem hiding this comment.
Pull request overview
Updates agent initialization to default to gpt-5.4-mini instead of deprecated gpt-4.1-mini.
Changes:
- Updates From Code and adopt/catalog model defaults.
- Refreshes model flag help and resolution guidance.
- Preserves explicit
--modeloverrides.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
init.go |
Updates --model help text. |
init_foundry_resources_helpers.go |
Changes the shared model default and error hint. |
init_adopt.go |
Changes the adopt/catalog model default. |
| fmt.Sprintf( | ||
| "pass --model <name> (e.g. --model %s) or --project-id "+ | ||
| "<id> with --model-deployment <name> to skip interactive model selection", | ||
| defaultAgentModel, | ||
| ), |
| "Name of the AI model to deploy. Defaults to '%s' during interactive model selection; "+ | ||
| "required to deploy a new model with --no-prompt. Mutually exclusive with --model-deployment, "+ | ||
| "with --model-deployment being used if both are provided", | ||
| defaultAgentModel, |
jongio
left a comment
jongio
left a comment
There was a problem hiding this comment.
Confirmed the deprecated default is centralized in defaultAgentModel and referenced from all three selection paths (From Code, adopt/catalog, and the shared helper) plus the --model help and the model-resolution error guidance, so the value can't drift out of sync. TestSelectNewModel exercises both the interactive default and the explicit --model override. Resolves #9119.
jongio
left a comment
jongio
left a comment
There was a problem hiding this comment.
Scope is limited to replacing the deprecated gpt-4.1-mini default with the shared defaultAgentModel constant across the From Code flow, adopt flow, help text, and error guidance. TestSelectNewModel covers both the interactive default and explicit --model precedence, and CI is green across Linux, Windows, and Mac.
jongio
left a comment
jongio
left a comment
There was a problem hiding this comment.
Confirmed the deprecated default is centralized in the shared defaultAgentModel constant across all selection paths (From Code, adopt/catalog, and error guidance) and the --model help text, so the value stays in sync. TestSelectNewModel asserts the prompt request DefaultValue and the explicit-flag override. CI is green.
jongio
left a comment
jongio
left a comment
There was a problem hiding this comment.
Verified the deprecated default is centralized in the shared defaultAgentModel constant and referenced from every selection path: the --model help text, the adopt/catalog flow, selectNewModel, and the model-resolution error guidance, so the value can't drift out of sync. The remaining gpt-4.1-mini literals are unrelated test fixtures (filter-exclusion and deployment-name cases), not defaults. TestSelectNewModel covers both the interactive default and the explicit --model override. CI is green across Linux, Windows, and Mac. Resolves #9119.
Closes #9119.
What changed
gpt-4.1-minidefault in interactive model catalog prompts withgpt-5.4-mini.defaultAgentModelso the From Code flow, adopt flow, help text, and error guidance stay synchronized.--modelhelp: the default applies during interactive selection, while--modelis required to deploy a new model with--no-prompt.--modelprecedence.An explicit
--model <name>continues to override the interactive default.Why this change is needed
gpt-4.1-miniis deprecated, so it should no longer be preselected when users choose a new model from the catalog. Leaving it as the default steers users toward an obsolete model and can eventually leave the prompt without its intended selection as catalog availability changes.The non-interactive behavior is intentionally unchanged. In the From Code flow,
--no-promptskips new-model configuration unless--modelis provided; an existing deployment can still be selected with--project-idand--model-deployment.Why this approach
gpt-5.4-miniis a current Azure AI Foundry model and preserves the small, cost-efficient intent of the previous default. A shared constant is used instead of repeating the model name so future default changes cannot leave prompts, help text, and guidance out of sync. The focused gRPC test verifies both the default request value and explicit flag precedence.