@juspay/neurolink 8.26.0 → 8.27.0

package/dist/cli/loop/optionsSchema.js

@@ -65,5 +65,18 @@ export const textGenerationOptionsSchema = {
         type: "boolean",
         description: "Enable or disable automatic conversation summarization for this request.",
     },
+    thinking: {
+        type: "boolean",
+        description: "Enable extended thinking/reasoning capability.",
+    },
+    thinkingBudget: {
+        type: "number",
+        description: "Token budget for thinking (Anthropic models: 5000-100000).",
+    },
+    thinkingLevel: {
+        type: "string",
+        description: "Thinking level for Gemini 3 models: minimal, low, medium, high.",
+        allowedValues: ["minimal", "low", "medium", "high"],
+    },
 };
 //# sourceMappingURL=optionsSchema.js.map

package/dist/config/modelSpecificPrompts.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Model-specific prompt configurations and enhancement utilities
+ */
+import { isGemini3Model, isGemini25Model } from "../utils/modelDetection.js";
+export { isGemini3Model, isGemini25Model };
+export declare const MODEL_SPECIFIC_INSTRUCTIONS: Record<string, string>;
+export declare function getModelSpecificInstructions(model: string): string;
+export declare function enhancePromptForModel(basePrompt: string, model: string, _provider?: string): string;
+export declare function shouldEnhancePrompt(model: string): boolean;

package/dist/config/modelSpecificPrompts.js

@@ -0,0 +1,38 @@
+/**
+ * Model-specific prompt configurations and enhancement utilities
+ */
+import { isGemini3Model, isGemini25Model } from "../utils/modelDetection.js";
+// Re-export from modelDetection for backwards compatibility
+export { isGemini3Model, isGemini25Model };
+export const MODEL_SPECIFIC_INSTRUCTIONS = {
+    "gemini-3": `You have access to extended thinking capabilities. Use them for complex reasoning tasks that require deep analysis.`,
+    "gemini-2.5": `You support function calling and structured outputs. Format responses according to the requested schema when provided.`,
+    "gpt-4": `You are a helpful assistant with strong reasoning capabilities.`,
+    "claude-3": `You have extended thinking capabilities available when enabled. Use systematic reasoning for complex problems.`,
+    default: "",
+};
+export function getModelSpecificInstructions(model) {
+    if (isGemini3Model(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gemini-3"];
+    }
+    if (isGemini25Model(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gemini-2.5"];
+    }
+    if (/^gpt-4/i.test(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gpt-4"];
+    }
+    if (/^claude-3/i.test(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["claude-3"];
+    }
+    return MODEL_SPECIFIC_INSTRUCTIONS["default"];
+}
+export function enhancePromptForModel(basePrompt, model, _provider) {
+    const modelInstructions = getModelSpecificInstructions(model);
+    if (!modelInstructions) {
+        return basePrompt;
+    }
+    return `${modelInstructions}\n\n${basePrompt}`;
+}
+export function shouldEnhancePrompt(model) {
+    return isGemini3Model(model) || isGemini25Model(model);
+}

package/dist/constants/enums.d.ts

@@ -191,6 +191,12 @@ export declare enum VertexModels {
     GEMINI_3_PRO_LATEST = "gemini-3-pro-latest",
     /** Gemini 3 Pro Preview - Generic preview (legacy) */
     GEMINI_3_PRO_PREVIEW = "gemini-3-pro-preview",
+    /** Gemini 3 Flash - Base model with adaptive thinking */
+    GEMINI_3_FLASH = "gemini-3-flash",
+    /** Gemini 3 Flash Preview - Versioned preview */
+    GEMINI_3_FLASH_PREVIEW = "gemini-3-flash-preview",
+    /** Gemini 3 Flash Latest - Auto-updated alias (always points to latest preview) */
+    GEMINI_3_FLASH_LATEST = "gemini-3-flash-latest",
     GEMINI_2_5_PRO = "gemini-2.5-pro",
     GEMINI_2_5_FLASH = "gemini-2.5-flash",
     GEMINI_2_5_FLASH_LITE = "gemini-2.5-flash-lite",
@@ -208,6 +214,8 @@ export declare enum VertexModels {
 export declare enum GoogleAIModels {
     GEMINI_3_PRO_PREVIEW = "gemini-3-pro-preview",
     GEMINI_3_PRO_IMAGE_PREVIEW = "gemini-3-pro-image-preview",
+    GEMINI_3_FLASH = "gemini-3-flash",
+    GEMINI_3_FLASH_PREVIEW = "gemini-3-flash-preview",
     GEMINI_2_5_PRO = "gemini-2.5-pro",
     GEMINI_2_5_FLASH = "gemini-2.5-flash",
     GEMINI_2_5_FLASH_LITE = "gemini-2.5-flash-lite",

package/dist/constants/enums.js

@@ -269,6 +269,12 @@ export var VertexModels;
     VertexModels["GEMINI_3_PRO_LATEST"] = "gemini-3-pro-latest";
     /** Gemini 3 Pro Preview - Generic preview (legacy) */
     VertexModels["GEMINI_3_PRO_PREVIEW"] = "gemini-3-pro-preview";
+    /** Gemini 3 Flash - Base model with adaptive thinking */
+    VertexModels["GEMINI_3_FLASH"] = "gemini-3-flash";
+    /** Gemini 3 Flash Preview - Versioned preview */
+    VertexModels["GEMINI_3_FLASH_PREVIEW"] = "gemini-3-flash-preview";
+    /** Gemini 3 Flash Latest - Auto-updated alias (always points to latest preview) */
+    VertexModels["GEMINI_3_FLASH_LATEST"] = "gemini-3-flash-latest";
     // Gemini 2.5 Series (Latest - 2025)
     VertexModels["GEMINI_2_5_PRO"] = "gemini-2.5-pro";
     VertexModels["GEMINI_2_5_FLASH"] = "gemini-2.5-flash";
@@ -291,6 +297,8 @@ export var GoogleAIModels;
     // Gemini 3 Series
     GoogleAIModels["GEMINI_3_PRO_PREVIEW"] = "gemini-3-pro-preview";
     GoogleAIModels["GEMINI_3_PRO_IMAGE_PREVIEW"] = "gemini-3-pro-image-preview";
+    GoogleAIModels["GEMINI_3_FLASH"] = "gemini-3-flash";
+    GoogleAIModels["GEMINI_3_FLASH_PREVIEW"] = "gemini-3-flash-preview";
     // Gemini 2.5 Series
     GoogleAIModels["GEMINI_2_5_PRO"] = "gemini-2.5-pro";
     GoogleAIModels["GEMINI_2_5_FLASH"] = "gemini-2.5-flash";

package/dist/constants/tokens.d.ts

@@ -85,6 +85,9 @@ export declare const PROVIDER_TOKEN_LIMITS: {
         readonly "gemini-3-pro-preview": 8192;
         readonly "gemini-3-pro-preview-11-2025": 8192;
         readonly "gemini-3-pro-latest": 8192;
+        readonly "gemini-3-flash": 65536;
+        readonly "gemini-3-flash-preview": 65536;
+        readonly "gemini-3-flash-latest": 65536;
         readonly "gemini-2.5-pro": 8192;
         readonly "gemini-2.5-flash": 8192;
         readonly "gemini-2.5-flash-lite": 8192;
@@ -100,6 +103,9 @@ export declare const PROVIDER_TOKEN_LIMITS: {
         readonly "gemini-3-pro-preview": 8192;
         readonly "gemini-3-pro-preview-11-2025": 8192;
         readonly "gemini-3-pro-latest": 8192;
+        readonly "gemini-3-flash": 65536;
+        readonly "gemini-3-flash-preview": 65536;
+        readonly "gemini-3-flash-latest": 65536;
         readonly "gemini-2.5-pro": 8192;
         readonly "gemini-2.5-flash": 8192;
         readonly "gemini-2.5-flash-lite": 8192;
@@ -182,6 +188,25 @@ export declare const CONTEXT_WINDOWS: {
     /** Maximum theoretical context */
     readonly MAXIMUM: 2097152;
 };
+/**
+ * Model-specific token limits with input/output breakdown
+ * For models that require explicit input and output token limits
+ */
+export declare const MODEL_TOKEN_LIMITS: {
+    /** Gemini 3 Flash Series */
+    readonly "gemini-3-flash": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+    readonly "gemini-3-flash-preview": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+    readonly "gemini-3-flash-latest": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+};
 /**
  * Token estimation utilities
  * Rough estimates for token counting without full tokenization

package/dist/constants/tokens.js

@@ -89,6 +89,10 @@ export const PROVIDER_TOKEN_LIMITS = {
         "gemini-3-pro-preview": 8192,
         "gemini-3-pro-preview-11-2025": 8192,
         "gemini-3-pro-latest": 8192,
+        // Gemini 3 Flash Series
+        "gemini-3-flash": 65536,
+        "gemini-3-flash-preview": 65536,
+        "gemini-3-flash-latest": 65536,
         // Gemini 2.5 Series
         "gemini-2.5-pro": 8192,
         "gemini-2.5-flash": 8192,
@@ -108,6 +112,10 @@ export const PROVIDER_TOKEN_LIMITS = {
         "gemini-3-pro-preview": 8192,
         "gemini-3-pro-preview-11-2025": 8192,
         "gemini-3-pro-latest": 8192,
+        // Gemini 3 Flash Series
+        "gemini-3-flash": 65536,
+        "gemini-3-flash-preview": 65536,
+        "gemini-3-flash-latest": 65536,
         // Gemini 2.5 Series
         "gemini-2.5-pro": 8192,
         "gemini-2.5-flash": 8192,
@@ -201,6 +209,16 @@ export const CONTEXT_WINDOWS = {
     /** Maximum theoretical context */
     MAXIMUM: 2097152, // 2M - Maximum context
 };
+/**
+ * Model-specific token limits with input/output breakdown
+ * For models that require explicit input and output token limits
+ */
+export const MODEL_TOKEN_LIMITS = {
+    /** Gemini 3 Flash Series */
+    "gemini-3-flash": { input: 1000000, output: 65536 },
+    "gemini-3-flash-preview": { input: 1000000, output: 65536 },
+    "gemini-3-flash-latest": { input: 1000000, output: 65536 },
+};
 /**
  * Token estimation utilities
  * Rough estimates for token counting without full tokenization

package/dist/core/analytics.js

@@ -6,6 +6,7 @@
  */
 import { logger } from "../utils/logger.js";
 import { modelConfig } from "./modelConfiguration.js";
+import { extractTokenUsage as extractTokenUsageUtil } from "../utils/tokenUtils.js";
 /**
  * Create analytics data structure from AI response
  */
@@ -49,36 +50,14 @@ export function createAnalytics(provider, model, result, responseTime, context)
 }
 /**
  * Extract token usage from various AI result formats
+ * Delegates to centralized tokenUtils for consistent extraction across providers
  */
 function extractTokenUsage(result) {
-    // Use properly typed usage object from BaseProvider or direct AI SDK
-    if (result.usage &&
-        typeof result.usage === "object" &&
-        result.usage !== null) {
-        const usage = result.usage;
-        // Try BaseProvider normalized format first (input/output/total)
-        if (typeof usage.input === "number" || typeof usage.output === "number") {
-            const input = typeof usage.input === "number" ? usage.input : 0;
-            const output = typeof usage.output === "number" ? usage.output : 0;
-            const total = typeof usage.total === "number" ? usage.total : input + output;
-            return { input, output, total };
-        }
-        // Try OpenAI/Mistral format (promptTokens/completionTokens)
-        if (typeof usage.promptTokens === "number" ||
-            typeof usage.completionTokens === "number") {
-            const input = typeof usage.promptTokens === "number" ? usage.promptTokens : 0;
-            const output = typeof usage.completionTokens === "number" ? usage.completionTokens : 0;
-            const total = typeof usage.total === "number" ? usage.total : input + output;
-            return { input, output, total };
-        }
-        // Handle total-only case
-        if (typeof usage.total === "number") {
-            return { input: 0, output: 0, total: usage.total };
-        }
-    }
-    // Fallback for edge cases
-    logger.debug("Token extraction failed: unknown usage format", { result });
-    return { input: 0, output: 0, total: 0 };
+    // Use centralized token extraction utility
+    // The utility handles nested usage objects, multiple provider formats,
+    // cache tokens, reasoning tokens, and cache savings calculation
+    // Cast result to allow extractTokenUsageUtil to handle type normalization
+    return extractTokenUsageUtil(result.usage);
 }
 /**
  * Estimate cost based on provider, model, and token usage

package/dist/core/baseProvider.js

@@ -126,6 +126,7 @@ export class BaseProvider {
                         maxSteps: options.maxSteps || 5,
                         provider: options.provider,
                         model: options.model,
+                        region: options.region, // Pass region for Vertex AI
                         // 🔧 FIX: Include analytics and evaluation options from stream options
                         enableAnalytics: options.enableAnalytics,
                         enableEvaluation: options.enableEvaluation,

package/dist/core/constants.d.ts

@@ -6,6 +6,7 @@ export declare const DEFAULT_MAX_TOKENS: undefined;
 export declare const DEFAULT_TEMPERATURE = 0.7;
 export declare const DEFAULT_TIMEOUT = 60000;
 export declare const DEFAULT_MAX_STEPS = 200;
+export declare const DEFAULT_TOOL_MAX_RETRIES = 2;
 export declare const STEP_LIMITS: {
     min: number;
     max: number;

package/dist/core/constants.js

@@ -7,6 +7,7 @@ export const DEFAULT_MAX_TOKENS = undefined; // Unlimited by default - let provi
 export const DEFAULT_TEMPERATURE = 0.7;
 export const DEFAULT_TIMEOUT = 60000;
 export const DEFAULT_MAX_STEPS = 200;
+export const DEFAULT_TOOL_MAX_RETRIES = 2; // Maximum retries per tool before permanently failing
 // Step execution limits
 export const STEP_LIMITS = {
     min: 1,

package/dist/core/modules/GenerationHandler.js

@@ -14,6 +14,7 @@
  */
 import { generateText, Output, NoObjectGeneratedError } from "ai";
 import { logger } from "../../utils/logger.js";
+import { extractTokenUsage } from "../../utils/tokenUtils.js";
 import { DEFAULT_MAX_STEPS } from "../constants.js";
 /**
  * GenerationHandler class - Handles text generation operations for AI providers
@@ -36,6 +37,10 @@ export class GenerationHandler {
      * @private
      */
     async callGenerateText(model, messages, tools, options, shouldUseTools, includeStructuredOutput) {
+        // Check if this is a Google provider (for provider-specific options)
+        const isGoogleProvider = this.providerName === "google-ai" || this.providerName === "vertex";
+        // Check if this is an Anthropic provider
+        const isAnthropicProvider = this.providerName === "anthropic" || this.providerName === "bedrock";
         const useStructuredOutput = includeStructuredOutput &&
             !!options.schema &&
             (options.output?.format === "json" ||
@@ -52,6 +57,39 @@ export class GenerationHandler {
                 options.schema && {
                 experimental_output: Output.object({ schema: options.schema }),
             }),
+            // Add thinking configuration for extended reasoning
+            // Gemini 3 models use providerOptions.google.thinkingConfig with thinkingLevel
+            // Gemini 2.5 models use thinkingBudget
+            // Anthropic models use experimental_thinking with budgetTokens
+            ...(options.thinkingConfig?.enabled && {
+                // For Anthropic: experimental_thinking with budgetTokens
+                ...(isAnthropicProvider &&
+                    options.thinkingConfig.budgetTokens &&
+                    !options.thinkingConfig.thinkingLevel && {
+                    experimental_thinking: {
+                        type: "enabled",
+                        budgetTokens: options.thinkingConfig.budgetTokens,
+                    },
+                }),
+                // For Google Gemini 3: providerOptions with thinkingLevel
+                // For Gemini 2.5: providerOptions with thinkingBudget
+                ...(isGoogleProvider && {
+                    providerOptions: {
+                        google: {
+                            thinkingConfig: {
+                                ...(options.thinkingConfig.thinkingLevel && {
+                                    thinkingLevel: options.thinkingConfig.thinkingLevel,
+                                }),
+                                ...(options.thinkingConfig.budgetTokens &&
+                                    !options.thinkingConfig.thinkingLevel && {
+                                    thinkingBudget: options.thinkingConfig.budgetTokens,
+                                }),
+                                includeThoughts: true,
+                            },
+                        },
+                    },
+                }),
+            }),
             experimental_telemetry: this.getTelemetryConfigFn(options, "generate"),
             onStepFinish: ({ toolCalls, toolResults }) => {
                 logger.info("Tool execution completed", { toolResults, toolCalls });
@@ -227,13 +265,13 @@ export class GenerationHandler {
         else {
             content = generateResult.text;
         }
+        // Extract usage with support for different formats and reasoning tokens
+        // Note: The AI SDK bundles thinking tokens into promptTokens for Google models.
+        // Separate reasoningTokens tracking will work when/if the AI SDK adds support.
+        const usage = extractTokenUsage(generateResult.usage);
         return {
             content,
-            usage: {
-                input: generateResult.usage?.promptTokens || 0,
-                output: generateResult.usage?.completionTokens || 0,
-                total: generateResult.usage?.totalTokens || 0,
-            },
+            usage,
             provider: this.providerName,
             model: this.modelName,
             toolCalls: generateResult.toolCalls

package/dist/core/streamAnalytics.d.ts

@@ -6,6 +6,7 @@ import type { StreamTextResult, StreamAnalyticsCollector, ResponseMetadata } fro
 export declare class BaseStreamAnalyticsCollector implements StreamAnalyticsCollector {
     /**
      * Collect token usage from stream result
+     * Uses centralized tokenUtils for consistent extraction across providers
      */
     collectUsage(result: StreamTextResult): Promise<TokenUsage>;
     /**

package/dist/core/streamAnalytics.js

@@ -1,37 +1,29 @@
 import { createAnalytics } from "./analytics.js";
 import { logger } from "../utils/logger.js";
+import { extractTokenUsage, createEmptyTokenUsage, } from "../utils/tokenUtils.js";
 /**
  * Base implementation for collecting analytics from Vercel AI SDK stream results
  */
 export class BaseStreamAnalyticsCollector {
     /**
      * Collect token usage from stream result
+     * Uses centralized tokenUtils for consistent extraction across providers
      */
     async collectUsage(result) {
         try {
             const usage = await result.usage;
             if (!usage) {
                 logger.debug("No usage data available from stream result");
-                return {
-                    input: 0,
-                    output: 0,
-                    total: 0,
-                };
+                return createEmptyTokenUsage();
             }
-            return {
-                input: usage.promptTokens || 0,
-                output: usage.completionTokens || 0,
-                total: usage.totalTokens ||
-                    (usage.promptTokens || 0) + (usage.completionTokens || 0),
-            };
+            // Use centralized token extraction utility
+            // Handles multiple provider formats, cache tokens, reasoning tokens,
+            // and cache savings calculation
+            return extractTokenUsage(usage);
         }
         catch (error) {
             logger.warn("Failed to collect usage from stream result", { error });
-            return {
-                input: 0,
-                output: 0,
-                total: 0,
-            };
+            return createEmptyTokenUsage();
         }
     }
     /**

package/dist/lib/adapters/providerImageAdapter.js

@@ -77,6 +77,10 @@ const VISION_CAPABILITIES = {
         "gemini-3-pro-preview-11-2025",
         "gemini-3-pro-latest",
         "gemini-3-pro-image-preview",
+        // Gemini 3 Flash Series
+        "gemini-3-flash",
+        "gemini-3-flash-preview",
+        "gemini-3-flash-latest",
         // Gemini 2.5 Series
         "gemini-2.5-pro",
         "gemini-2.5-flash",
@@ -151,6 +155,10 @@ const VISION_CAPABILITIES = {
         "gemini-3-pro-latest",
         "gemini-3-pro-preview",
         "gemini-3-pro",
+        // Gemini 3 Flash Series on Vertex AI
+        "gemini-3-flash",
+        "gemini-3-flash-preview",
+        "gemini-3-flash-latest",
         // Gemini 2.5 models on Vertex AI
         "gemini-2.5-pro",
         "gemini-2.5-flash",
@@ -226,6 +234,9 @@ const VISION_CAPABILITIES = {
         "gemini/gemini-2.0-flash",
         "gemini-3-pro-preview",
         "gemini-3-pro-latest",
+        "gemini-3-flash",
+        "gemini-3-flash-preview",
+        "gemini-3-flash-latest",
         "gemini-2.5-pro",
         "gemini-2.5-flash",
         "gemini-2.0-flash-lite",

package/dist/lib/config/modelSpecificPrompts.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Model-specific prompt configurations and enhancement utilities
+ */
+import { isGemini3Model, isGemini25Model } from "../utils/modelDetection.js";
+export { isGemini3Model, isGemini25Model };
+export declare const MODEL_SPECIFIC_INSTRUCTIONS: Record<string, string>;
+export declare function getModelSpecificInstructions(model: string): string;
+export declare function enhancePromptForModel(basePrompt: string, model: string, _provider?: string): string;
+export declare function shouldEnhancePrompt(model: string): boolean;

package/dist/lib/config/modelSpecificPrompts.js

@@ -0,0 +1,39 @@
+/**
+ * Model-specific prompt configurations and enhancement utilities
+ */
+import { isGemini3Model, isGemini25Model } from "../utils/modelDetection.js";
+// Re-export from modelDetection for backwards compatibility
+export { isGemini3Model, isGemini25Model };
+export const MODEL_SPECIFIC_INSTRUCTIONS = {
+    "gemini-3": `You have access to extended thinking capabilities. Use them for complex reasoning tasks that require deep analysis.`,
+    "gemini-2.5": `You support function calling and structured outputs. Format responses according to the requested schema when provided.`,
+    "gpt-4": `You are a helpful assistant with strong reasoning capabilities.`,
+    "claude-3": `You have extended thinking capabilities available when enabled. Use systematic reasoning for complex problems.`,
+    default: "",
+};
+export function getModelSpecificInstructions(model) {
+    if (isGemini3Model(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gemini-3"];
+    }
+    if (isGemini25Model(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gemini-2.5"];
+    }
+    if (/^gpt-4/i.test(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["gpt-4"];
+    }
+    if (/^claude-3/i.test(model)) {
+        return MODEL_SPECIFIC_INSTRUCTIONS["claude-3"];
+    }
+    return MODEL_SPECIFIC_INSTRUCTIONS["default"];
+}
+export function enhancePromptForModel(basePrompt, model, _provider) {
+    const modelInstructions = getModelSpecificInstructions(model);
+    if (!modelInstructions) {
+        return basePrompt;
+    }
+    return `${modelInstructions}\n\n${basePrompt}`;
+}
+export function shouldEnhancePrompt(model) {
+    return isGemini3Model(model) || isGemini25Model(model);
+}
+//# sourceMappingURL=modelSpecificPrompts.js.map

package/dist/lib/constants/enums.d.ts

@@ -191,6 +191,12 @@ export declare enum VertexModels {
     GEMINI_3_PRO_LATEST = "gemini-3-pro-latest",
     /** Gemini 3 Pro Preview - Generic preview (legacy) */
     GEMINI_3_PRO_PREVIEW = "gemini-3-pro-preview",
+    /** Gemini 3 Flash - Base model with adaptive thinking */
+    GEMINI_3_FLASH = "gemini-3-flash",
+    /** Gemini 3 Flash Preview - Versioned preview */
+    GEMINI_3_FLASH_PREVIEW = "gemini-3-flash-preview",
+    /** Gemini 3 Flash Latest - Auto-updated alias (always points to latest preview) */
+    GEMINI_3_FLASH_LATEST = "gemini-3-flash-latest",
     GEMINI_2_5_PRO = "gemini-2.5-pro",
     GEMINI_2_5_FLASH = "gemini-2.5-flash",
     GEMINI_2_5_FLASH_LITE = "gemini-2.5-flash-lite",
@@ -208,6 +214,8 @@ export declare enum VertexModels {
 export declare enum GoogleAIModels {
     GEMINI_3_PRO_PREVIEW = "gemini-3-pro-preview",
     GEMINI_3_PRO_IMAGE_PREVIEW = "gemini-3-pro-image-preview",
+    GEMINI_3_FLASH = "gemini-3-flash",
+    GEMINI_3_FLASH_PREVIEW = "gemini-3-flash-preview",
     GEMINI_2_5_PRO = "gemini-2.5-pro",
     GEMINI_2_5_FLASH = "gemini-2.5-flash",
     GEMINI_2_5_FLASH_LITE = "gemini-2.5-flash-lite",

package/dist/lib/constants/enums.js

@@ -269,6 +269,12 @@ export var VertexModels;
     VertexModels["GEMINI_3_PRO_LATEST"] = "gemini-3-pro-latest";
     /** Gemini 3 Pro Preview - Generic preview (legacy) */
     VertexModels["GEMINI_3_PRO_PREVIEW"] = "gemini-3-pro-preview";
+    /** Gemini 3 Flash - Base model with adaptive thinking */
+    VertexModels["GEMINI_3_FLASH"] = "gemini-3-flash";
+    /** Gemini 3 Flash Preview - Versioned preview */
+    VertexModels["GEMINI_3_FLASH_PREVIEW"] = "gemini-3-flash-preview";
+    /** Gemini 3 Flash Latest - Auto-updated alias (always points to latest preview) */
+    VertexModels["GEMINI_3_FLASH_LATEST"] = "gemini-3-flash-latest";
     // Gemini 2.5 Series (Latest - 2025)
     VertexModels["GEMINI_2_5_PRO"] = "gemini-2.5-pro";
     VertexModels["GEMINI_2_5_FLASH"] = "gemini-2.5-flash";
@@ -291,6 +297,8 @@ export var GoogleAIModels;
     // Gemini 3 Series
     GoogleAIModels["GEMINI_3_PRO_PREVIEW"] = "gemini-3-pro-preview";
     GoogleAIModels["GEMINI_3_PRO_IMAGE_PREVIEW"] = "gemini-3-pro-image-preview";
+    GoogleAIModels["GEMINI_3_FLASH"] = "gemini-3-flash";
+    GoogleAIModels["GEMINI_3_FLASH_PREVIEW"] = "gemini-3-flash-preview";
     // Gemini 2.5 Series
     GoogleAIModels["GEMINI_2_5_PRO"] = "gemini-2.5-pro";
     GoogleAIModels["GEMINI_2_5_FLASH"] = "gemini-2.5-flash";

package/dist/lib/constants/tokens.d.ts

@@ -85,6 +85,9 @@ export declare const PROVIDER_TOKEN_LIMITS: {
         readonly "gemini-3-pro-preview": 8192;
         readonly "gemini-3-pro-preview-11-2025": 8192;
         readonly "gemini-3-pro-latest": 8192;
+        readonly "gemini-3-flash": 65536;
+        readonly "gemini-3-flash-preview": 65536;
+        readonly "gemini-3-flash-latest": 65536;
         readonly "gemini-2.5-pro": 8192;
         readonly "gemini-2.5-flash": 8192;
         readonly "gemini-2.5-flash-lite": 8192;
@@ -100,6 +103,9 @@ export declare const PROVIDER_TOKEN_LIMITS: {
         readonly "gemini-3-pro-preview": 8192;
         readonly "gemini-3-pro-preview-11-2025": 8192;
         readonly "gemini-3-pro-latest": 8192;
+        readonly "gemini-3-flash": 65536;
+        readonly "gemini-3-flash-preview": 65536;
+        readonly "gemini-3-flash-latest": 65536;
         readonly "gemini-2.5-pro": 8192;
         readonly "gemini-2.5-flash": 8192;
         readonly "gemini-2.5-flash-lite": 8192;
@@ -182,6 +188,25 @@ export declare const CONTEXT_WINDOWS: {
     /** Maximum theoretical context */
     readonly MAXIMUM: 2097152;
 };
+/**
+ * Model-specific token limits with input/output breakdown
+ * For models that require explicit input and output token limits
+ */
+export declare const MODEL_TOKEN_LIMITS: {
+    /** Gemini 3 Flash Series */
+    readonly "gemini-3-flash": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+    readonly "gemini-3-flash-preview": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+    readonly "gemini-3-flash-latest": {
+        readonly input: 1000000;
+        readonly output: 65536;
+    };
+};
 /**
  * Token estimation utilities
  * Rough estimates for token counting without full tokenization

package/dist/lib/constants/tokens.js

@@ -89,6 +89,10 @@ export const PROVIDER_TOKEN_LIMITS = {
         "gemini-3-pro-preview": 8192,
         "gemini-3-pro-preview-11-2025": 8192,
         "gemini-3-pro-latest": 8192,
+        // Gemini 3 Flash Series
+        "gemini-3-flash": 65536,
+        "gemini-3-flash-preview": 65536,
+        "gemini-3-flash-latest": 65536,
         // Gemini 2.5 Series
         "gemini-2.5-pro": 8192,
         "gemini-2.5-flash": 8192,
@@ -108,6 +112,10 @@ export const PROVIDER_TOKEN_LIMITS = {
         "gemini-3-pro-preview": 8192,
         "gemini-3-pro-preview-11-2025": 8192,
         "gemini-3-pro-latest": 8192,
+        // Gemini 3 Flash Series
+        "gemini-3-flash": 65536,
+        "gemini-3-flash-preview": 65536,
+        "gemini-3-flash-latest": 65536,
         // Gemini 2.5 Series
         "gemini-2.5-pro": 8192,
         "gemini-2.5-flash": 8192,
@@ -201,6 +209,16 @@ export const CONTEXT_WINDOWS = {
     /** Maximum theoretical context */
     MAXIMUM: 2097152, // 2M - Maximum context
 };
+/**
+ * Model-specific token limits with input/output breakdown
+ * For models that require explicit input and output token limits
+ */
+export const MODEL_TOKEN_LIMITS = {
+    /** Gemini 3 Flash Series */
+    "gemini-3-flash": { input: 1000000, output: 65536 },
+    "gemini-3-flash-preview": { input: 1000000, output: 65536 },
+    "gemini-3-flash-latest": { input: 1000000, output: 65536 },
+};
 /**
  * Token estimation utilities
  * Rough estimates for token counting without full tokenization