Move the AI Answer default off gpt-3.5-turbo before its 23 October shutdown

plugins/openai/body_test.go

@@ -0,0 +1,30 @@
+package openai
+
+import "testing"
+
+func TestBuildChatGPTBodyTokenLimitField(t *testing.T) {
+	limit := 300
+	messages := []map[string]interface{}{{"role": "user", "content": "hi"}}
+
+	body := BuildChatGPTBody("gpt-5.6-terra", messages, &limit, nil, false, OpenAIType)
+	if _, ok := body["max_tokens"]; ok {
+		t.Errorf("gpt-5 body must not send max_tokens: %v", body)
+	}
+	if body["max_completion_tokens"] != 300 {
+		t.Errorf("gpt-5 body must send max_completion_tokens=300: %v", body)
+	}
+
+	body = BuildChatGPTBody("gpt-4.1-mini", messages, &limit, nil, false, OpenAIType)
+	if body["max_tokens"] != 300 {
+		t.Errorf("gpt-4.1-mini body must keep max_tokens=300: %v", body)
+	}
+	if _, ok := body["max_completion_tokens"]; ok {
+		t.Errorf("gpt-4.1-mini body must not send max_completion_tokens: %v", body)
+	}
+}
+
+func TestDefaultModelIsNotRetiring(t *testing.T) {
+	if defaultModel != "gpt-4.1-mini" {
+		t.Errorf("defaultModel = %q, want gpt-4.1-mini (gpt-3.5-turbo shuts down 23 Oct 2026)", defaultModel)
+	}
+}

plugins/openai/tokens_test.go

@@ -210,6 +210,13 @@ func TestTrimContextAsPerLimits(t *testing.T) {
 	json.Unmarshal(messagesAsBytes, &messagesAsMap)
 
 	openAIInstance := Instance()
+	// These cases are sized for a 4,096-token context window. The default
+	// model is no longer gpt-3.5-turbo, so name it explicitly.
+	previous := openAIInstance.GetConfig()
+	model := "gpt-3.5-turbo"
+	openAIInstance.SetConfig(OpenAIConfig{Model: &model})
+	defer openAIInstance.SetConfig(previous)
+
 	Convey("Within Limit messages 1 trimmed", t, func() {
 		messagesToPass := messagesAsMap
 		updatedMessages, maxTokensToUse, err := openAIInstance.TrimContextAsPerLimits(messagesToPass, 2048, 3700, false)

plugins/openai/util.go

@@ -27,7 +27,7 @@ var (
 
 const (
 	OpenAIAPIURL     string = "https://api.openai.com/v1"
-	defaultModel     string = "gpt-3.5-turbo"
+	defaultModel     string = "gpt-4.1-mini"
 	defaultPrompt    string = "You're a helpful assistant."
 	defaultMaxTokens int    = 300
 	defaultMinTokens int    = 100
@@ -830,7 +830,7 @@ func (o OpenAIConfig) Key() string {
 // call.
 //
 // This will prioritize the model that user has configured, if
-// any and fallback to `gpt-3.5-turbo`
+// any and fallback to `gpt-4.1-mini`
 func (o OpenAIConfig) GetModel() string {
 	modelSet := o.Model
 	if modelSet == nil {
@@ -1262,9 +1262,14 @@ func BuildChatGPTBody(model string, messages []map[string]interface{}, maxTokens
 		requestBodyAsMap["model"] = model
 	}
 
-	// If maxTokens is passed, inject it in the request body
+	// If maxTokens is passed, inject it in the request body. GPT-5 models
+	// reject `max_tokens` and take the limit as `max_completion_tokens`.
 	if maxTokens != nil {
-		requestBodyAsMap["max_tokens"] = *maxTokens
+		if strings.HasPrefix(model, "gpt-5") {
+			requestBodyAsMap["max_completion_tokens"] = *maxTokens
+		} else {
+			requestBodyAsMap["max_tokens"] = *maxTokens
+		}
 	}
 
 	// If temperature is passed, inject it in the request body

plugins/pipelines/ai_answer.go

@@ -106,7 +106,7 @@ func executeAIAnswerStage(
 
 	// Set default model if none is passed
 	if inputs.Model == nil {
-		defaultModel := "gpt-3.5-turbo"
+		defaultModel := "gpt-4.1-mini"
 		inputs.Model = &defaultModel
 	}