@f5-sales-demo/pi-ai 21.0.0 → 21.2.0

package/CHANGELOG.md

@@ -4,6 +4,10 @@
 
 ## Unreleased
 
+### Changed
+
+- 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
 
 - Omit an empty OpenAI-compatible `tools` field unless prior tool-call history requires the explicit empty array, and suppress the internal no-auth credential sentinel from keyless request headers ([#2942](https://github.com/f5-sales-demo/xcsh/issues/2942), [#3361](https://github.com/f5-sales-demo/xcsh/issues/3361)).

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/pi-ai",
-	"version": "21.0.0",
+	"version": "21.2.0",
 	"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": "21.0.0",
+		"@f5-sales-demo/pi-utils": "21.2.0",
 		"@google/genai": "^2.15",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",

package/src/model-thinking.ts

@@ -80,7 +80,7 @@ const ANTHROPIC_ADAPTIVE_EFFORTS: readonly Effort[] = [
  */
 const ANTHROPIC_EXTENDED_EFFORT_VERSIONS: ReadonlySet<string> = new Set(["5.0"]);
 
-const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High];
+const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High];
 const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High];
 const GPT_5_2_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
 const GPT_5_1_CODEX_MINI_EFFORTS: readonly Effort[] = [Effort.Medium, Effort.High];
@@ -397,11 +397,22 @@ function inferModelThinking<TApi extends Api>(model: ApiModel<TApi>): ThinkingCo
 	if (!minLevel || !maxLevel) {
 		throw new Error(`Model ${model.provider}/${model.id} resolved to an empty thinking range`);
 	}
-	return {
+	const thinking: ThinkingConfig = {
 		mode: inferThinkingControlMode(model, parsedModel),
 		minLevel,
 		maxLevel,
 	};
+	if (
+		model.provider === "google-vertex" &&
+		parsedModel.family === "gemini" &&
+		!model.id.toLowerCase().includes("lite")
+	) {
+		thinking.defaultLevel = Effort.High;
+		if (parsedModel.version.major === 3) {
+			thinking.canDisable = false;
+		}
+	}
+	return thinking;
 }
 
 function normalizeThinkingConfig(thinking: ThinkingConfig | undefined): ThinkingConfig | undefined {
@@ -414,7 +425,13 @@ function normalizeThinkingConfig(thinking: ThinkingConfig | undefined): Thinking
 function thinkingsEqual(left: ThinkingConfig | undefined, right: ThinkingConfig | undefined): boolean {
 	if (left === right) return true;
 	if (!left || !right) return false;
-	return left.mode === right.mode && left.minLevel === right.minLevel && left.maxLevel === right.maxLevel;
+	return (
+		left.mode === right.mode &&
+		left.minLevel === right.minLevel &&
+		left.maxLevel === right.maxLevel &&
+		left.defaultLevel === right.defaultLevel &&
+		left.canDisable === right.canDisable
+	);
 }
 
 function expandEffortRange(thinking: ThinkingConfig): readonly Effort[] {

package/src/models.json

@@ -7030,9 +7030,9 @@
 		}
 	},
 	"google-vertex": {
-		"gemini-3.6-flash": {
-			"id": "gemini-3.6-flash",
-			"name": "Gemini 3.6 Flash",
+		"gemini-3.7-flash": {
+			"id": "gemini-3.7-flash",
+			"name": "Gemini 3.7 Flash",
 			"api": "google-vertex",
 			"provider": "google-vertex",
 			"baseUrl": "https://{location}-aiplatform.googleapis.com",
@@ -7042,9 +7042,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
+				"input": 0.75,
+				"output": 3.75,
+				"cacheRead": 0.075,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -7052,7 +7052,9 @@
 			"thinking": {
 				"mode": "google-level",
 				"minLevel": "low",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high",
+				"canDisable": false
 			}
 		},
 		"gemini-1.5-flash": {
@@ -7177,7 +7179,8 @@
 			"thinking": {
 				"mode": "budget",
 				"minLevel": "minimal",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high"
 			}
 		},
 		"gemini-2.5-flash-lite": {
@@ -7252,7 +7255,8 @@
 			"thinking": {
 				"mode": "budget",
 				"minLevel": "minimal",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high"
 			}
 		},
 		"gemini-3-flash-preview": {
@@ -7277,7 +7281,9 @@
 			"thinking": {
 				"mode": "google-level",
 				"minLevel": "minimal",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high",
+				"canDisable": false
 			}
 		},
 		"gemini-3-pro-preview": {
@@ -7302,7 +7308,9 @@
 			"thinking": {
 				"mode": "google-level",
 				"minLevel": "low",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high",
+				"canDisable": false
 			}
 		},
 		"gemini-3.1-pro-preview": {
@@ -7327,7 +7335,9 @@
 			"thinking": {
 				"mode": "google-level",
 				"minLevel": "low",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high",
+				"canDisable": false
 			}
 		},
 		"gemini-3.1-pro-preview-customtools": {
@@ -7352,7 +7362,9 @@
 			"thinking": {
 				"mode": "google-level",
 				"minLevel": "low",
-				"maxLevel": "high"
+				"maxLevel": "high",
+				"defaultLevel": "high",
+				"canDisable": false
 			}
 		}
 	},

package/src/providers/google-shared.ts

@@ -62,10 +62,10 @@ function resolveThoughtSignature(isSameProviderAndModel: boolean, signature: str
 }
 
 /**
- * Claude models via Google APIs require explicit tool call IDs in function calls/responses.
+ * Models with strict Google tool round trips require explicit matching call/response IDs.
  */
 export function requiresToolCallId(modelId: string): boolean {
-	return modelId.startsWith("claude-");
+	return modelId.startsWith("claude-") || modelId === "gemini-3.7-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" && part?.functionCall?.id) {
+					if (model.provider === "google-vertex" && model.id !== "gemini-3.7-flash" && part?.functionCall?.id) {
 						delete part.functionCall.id; // Vertex AI does not support 'id' in functionCall
 					}
 					if (effectiveSignature) {
@@ -228,7 +228,11 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 				},
 			};
 
-			if (model.provider === "google-vertex" && functionResponsePart.functionResponse?.id) {
+			if (
+				model.provider === "google-vertex" &&
+				model.id !== "gemini-3.7-flash" &&
+				functionResponsePart.functionResponse?.id
+			) {
 				delete functionResponsePart.functionResponse.id; // Vertex AI does not support 'id' in functionResponse
 			}
 

package/src/providers/google-vertex.ts

@@ -108,7 +108,7 @@ export const streamGoogleVertex: StreamFunction<"google-vertex"> = (
 			const project = apiKey ? undefined : await resolveGoogleVertexProject(options);
 			const location = apiKey ? undefined : resolveGoogleVertexLocation(options);
 			const client = apiKey ? createClientWithApiKey(model, apiKey) : createClient(model, project!, location!);
-			const params = buildParams(model, context, options);
+			const params = buildGoogleVertexParams(model, context, options);
 			options?.onPayload?.(params);
 			rawRequestDump = {
 				provider: model.provider,
@@ -467,33 +467,34 @@ export function googleVertexRequestUrl(modelId: string, project: string, locatio
 	return `https://${host}/${API_VERSION}/projects/${project}/locations/${location}/publishers/google/models/${modelId}:streamGenerateContent`;
 }
 
-function buildParams(
+export function buildGoogleVertexParams(
 	model: Model<"google-vertex">,
 	context: Context,
 	options: GoogleVertexOptions = {},
 ): GenerateContentParameters {
 	const contents = convertMessages(model, context);
+	const isGemini37Flash = model.id === "gemini-3.7-flash";
 
 	const generationConfig: GoogleVertexSamplingConfig = {};
-	if (options.temperature !== undefined) {
+	if (!isGemini37Flash && options.temperature !== undefined) {
 		generationConfig.temperature = options.temperature;
 	}
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
-	if (options.topP !== undefined) {
+	if (!isGemini37Flash && options.topP !== undefined) {
 		generationConfig.topP = options.topP;
 	}
-	if (options.topK !== undefined) {
+	if (!isGemini37Flash && options.topK !== undefined) {
 		generationConfig.topK = options.topK;
 	}
-	if (options.minP !== undefined) {
+	if (!isGemini37Flash && options.minP !== undefined) {
 		generationConfig.minP = options.minP;
 	}
-	if (options.presencePenalty !== undefined) {
+	if (!isGemini37Flash && options.presencePenalty !== undefined) {
 		generationConfig.presencePenalty = options.presencePenalty;
 	}
-	if (options.repetitionPenalty !== undefined) {
+	if (!isGemini37Flash && options.repetitionPenalty !== undefined) {
 		generationConfig.repetitionPenalty = options.repetitionPenalty;
 	}
 
@@ -513,11 +514,13 @@ function buildParams(
 		config.toolConfig = undefined;
 	}
 
-	if (options.thinking?.enabled && model.reasoning) {
+	if (model.reasoning && (options.thinking?.enabled || isGemini37Flash)) {
 		const cfg: ThinkingConfig = { includeThoughts: true };
-		if (options.thinking.level !== undefined) {
+		if (options.thinking?.level !== undefined) {
 			cfg.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];
-		} else if (options.thinking.budgetTokens !== undefined) {
+		} else if (isGemini37Flash) {
+			cfg.thinkingLevel = ThinkingLevel.HIGH;
+		} else if (options.thinking?.budgetTokens !== undefined) {
 			cfg.thinkingBudget = options.thinking.budgetTokens;
 		}
 		config.thinkingConfig = cfg;

package/src/types.ts

@@ -83,6 +83,10 @@ export interface ThinkingConfig {
 	minLevel: Effort;
 	/** Most intensive supported user-facing effort level. */
 	maxLevel: Effort;
+	/** Model-specific effort used when no explicit or saved effort is selected. */
+	defaultLevel?: Effort;
+	/** Whether callers may explicitly disable thinking. Defaults to true. */
+	canDisable?: boolean;
 	/** Provider-specific transport used to encode the selected effort. */
 	mode: ThinkingControlMode;
 }