@f5-sales-demo/pi-ai 21.16.0 → 21.18.0

package/CHANGELOG.md

@@ -6,6 +6,9 @@
 
 ### Changed
 
+- Replaced the bundled Vertex Gemini 3.7 Flash model and provider default with GA Gemini 3.8 Flash,
+  retaining HIGH thinking by default and the strict Vertex tool-call contract
+  ([#3751](https://github.com/f5-sales-demo/xcsh/issues/3751)).
 - Replaced Vertex Gemini 3.6 Flash with GA Gemini 3.7 Flash and made full Vertex Gemini Flash and Pro models default to HIGH thinking while preserving supported explicit effort overrides ([#3460](https://github.com/f5-sales-demo/xcsh/issues/3460)).
 
 ### Fixed

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/pi-ai",
-	"version": "21.16.0",
+	"version": "21.18.0",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/f5-sales-demo/xcsh",
 	"author": "Can Boluk",
@@ -46,7 +46,7 @@
 		"@anthropic-ai/sdk": "^0.115",
 		"@aws-sdk/client-bedrock-runtime": "^3",
 		"@bufbuild/protobuf": "^2.11",
-		"@f5-sales-demo/pi-utils": "21.16.0",
+		"@f5-sales-demo/pi-utils": "21.18.0",
 		"@google/genai": "^2.15",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",

package/src/models.json

@@ -10316,9 +10316,9 @@
 		}
 	},
 	"google-vertex": {
-		"gemini-3.7-flash": {
-			"id": "gemini-3.7-flash",
-			"name": "Gemini 3.7 Flash",
+		"gemini-3.8-flash": {
+			"id": "gemini-3.8-flash",
+			"name": "Gemini 3.8 Flash",
 			"api": "google-vertex",
 			"provider": "google-vertex",
 			"baseUrl": "https://{location}-aiplatform.googleapis.com",
@@ -10338,10 +10338,6 @@
 			"thinking": {
 				"mode": "google-level",
 				"supportedLevels": [
-					{
-						"effort": "minimal",
-						"description": "Very brief reasoning"
-					},
 					{
 						"effort": "low",
 						"description": "Light reasoning"

package/src/provider-models/descriptors.ts

@@ -276,7 +276,7 @@ export const DEFAULT_MODEL_PER_PROVIDER: Record<KnownProvider, string> = {
 	"amazon-bedrock": "us.anthropic.claude-opus-4-6-v1",
 	"google-antigravity": "gemini-3-pro-high",
 	"google-gemini-cli": "gemini-2.5-pro",
-	"google-vertex": "gemini-3-pro-preview",
+	"google-vertex": "gemini-3.8-flash",
 	minimax: "MiniMax-M2.5",
 	"minimax-code": "MiniMax-M2.5",
 	"minimax-code-cn": "MiniMax-M2.5",

package/src/providers/google-shared.ts

@@ -65,7 +65,7 @@ function resolveThoughtSignature(isSameProviderAndModel: boolean, signature: str
  * Models with strict Google tool round trips require explicit matching call/response IDs.
  */
 export function requiresToolCallId(modelId: string): boolean {
-	return modelId.startsWith("claude-") || modelId === "gemini-3.7-flash";
+	return modelId.startsWith("claude-") || modelId === "gemini-3.8-flash";
 }
 
 function getGeminiMajorVersion(modelId: string): number | undefined {
@@ -177,7 +177,7 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 							...(requiresToolCallId(model.id) ? { id: block.id } : {}),
 						},
 					};
-					if (model.provider === "google-vertex" && model.id !== "gemini-3.7-flash" && part?.functionCall?.id) {
+					if (model.provider === "google-vertex" && model.id !== "gemini-3.8-flash" && part?.functionCall?.id) {
 						delete part.functionCall.id; // Vertex AI does not support 'id' in functionCall
 					}
 					if (effectiveSignature) {
@@ -230,7 +230,7 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 
 			if (
 				model.provider === "google-vertex" &&
-				model.id !== "gemini-3.7-flash" &&
+				model.id !== "gemini-3.8-flash" &&
 				functionResponsePart.functionResponse?.id
 			) {
 				delete functionResponsePart.functionResponse.id; // Vertex AI does not support 'id' in functionResponse

package/src/providers/google-vertex.ts

@@ -450,7 +450,7 @@ export async function resolveGoogleVertexProject(
 }
 
 export function resolveGoogleVertexLocation(options?: GoogleVertexOptions): string {
-	// Gemini 3.7 Flash Corporate is served through the global endpoint only.
+	// Gemini 3.8 Flash Corporate is served through the global endpoint only.
 	// Ignore inherited provider/environment locations so a request cannot drift.
 	void options;
 	return "global";
@@ -484,28 +484,31 @@ export function buildGoogleVertexParams(
 	options: GoogleVertexOptions = {},
 ): GenerateContentParameters {
 	const contents = convertMessages(model, context);
-	const isGemini37Flash = model.id === "gemini-3.7-flash";
+	const isGemini38Flash = model.id === "gemini-3.8-flash";
+	if (isGemini38Flash && options.thinking?.level === "MINIMAL") {
+		throw new Error("MINIMAL thinking is not supported by Gemini 3.8 Flash");
+	}
 
 	const generationConfig: GoogleVertexSamplingConfig = {};
-	if (!isGemini37Flash && options.temperature !== undefined) {
+	if (!isGemini38Flash && options.temperature !== undefined) {
 		generationConfig.temperature = options.temperature;
 	}
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
-	if (!isGemini37Flash && options.topP !== undefined) {
+	if (!isGemini38Flash && options.topP !== undefined) {
 		generationConfig.topP = options.topP;
 	}
-	if (!isGemini37Flash && options.topK !== undefined) {
+	if (!isGemini38Flash && options.topK !== undefined) {
 		generationConfig.topK = options.topK;
 	}
-	if (!isGemini37Flash && options.minP !== undefined) {
+	if (!isGemini38Flash && options.minP !== undefined) {
 		generationConfig.minP = options.minP;
 	}
-	if (!isGemini37Flash && options.presencePenalty !== undefined) {
+	if (!isGemini38Flash && options.presencePenalty !== undefined) {
 		generationConfig.presencePenalty = options.presencePenalty;
 	}
-	if (!isGemini37Flash && options.repetitionPenalty !== undefined) {
+	if (!isGemini38Flash && options.repetitionPenalty !== undefined) {
 		generationConfig.repetitionPenalty = options.repetitionPenalty;
 	}
 
@@ -527,11 +530,11 @@ export function buildGoogleVertexParams(
 		config.toolConfig = undefined;
 	}
 
-	if (model.reasoning && (options.thinking?.enabled || isGemini37Flash)) {
+	if (model.reasoning && (options.thinking?.enabled || isGemini38Flash)) {
 		const cfg: ThinkingConfig = { includeThoughts: true };
 		if (options.thinking?.level !== undefined) {
 			cfg.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];
-		} else if (isGemini37Flash) {
+		} else if (isGemini38Flash) {
 			cfg.thinkingLevel = ThinkingLevel.HIGH;
 		} else if (options.thinking?.budgetTokens !== undefined) {
 			cfg.thinkingBudget = options.thinking.budgetTokens;