diff --git a/dist/main_bun.mjs b/dist/main_bun.mjs
index c5a2a92..77d9a81 100644
--- a/dist/main_bun.mjs
+++ b/dist/main_bun.mjs
@@ -102,13 +102,7 @@ function openAiMessageToGeminiMessage(messages) {
   return result;
 }
 function genModel(req) {
-  const defaultModel = (m) => {
-    if (m.startsWith("gemini")) {
-      return m;
-    }
-    return "gemini-1.5-flash-latest";
-  };
-  const model = ModelMapping[req.model] ?? defaultModel(req.model);
+  const model = GeminiModel.modelMapping(req.model);
   let functions = req.tools?.filter((it) => it.type === "function")?.map((it) => it.function) ?? [];
   functions = functions.concat((req.functions ?? []).map((it) => ({ strict: null, ...it })));
   const responseMimeType = req.response_format?.type === "json_object" ? "application/json" : "text/plain";
@@ -118,7 +112,10 @@ function genModel(req) {
       maxOutputTokens: req.max_tokens ?? void 0,
       temperature: req.temperature ?? void 0,
       topP: req.top_p ?? void 0,
-      responseMimeType
+      responseMimeType,
+      thinkingConfig: !model.isThinkingModel() ? void 0 : {
+        includeThoughts: true
+      }
     },
     tools: functions.length === 0 ? void 0 : [
       {
@@ -137,6 +134,34 @@ function genModel(req) {
   };
   return [model, generateContentRequest];
 }
+var GeminiModel = class _GeminiModel {
+  static modelMapping(model) {
+    const modelName = ModelMapping[model] ?? _GeminiModel.defaultModel(model);
+    return new _GeminiModel(modelName);
+  }
+  model;
+  constructor(model) {
+    this.model = model;
+  }
+  isThinkingModel() {
+    return this.model.includes("thinking");
+  }
+  apiVersion() {
+    if (this.isThinkingModel()) {
+      return "v1alpha";
+    }
+    return "v1beta";
+  }
+  toString() {
+    return this.model;
+  }
+  static defaultModel(m) {
+    if (m.startsWith("gemini")) {
+      return m;
+    }
+    return "gemini-1.5-flash-latest";
+  }
+};
 var ModelMapping = {
   "gpt-3.5-turbo": "gemini-1.5-flash-8b-latest",
   "gpt-4": "gemini-1.5-pro-latest",
@@ -405,7 +430,7 @@ var RequestUrl = class {
     this.apiParam = apiParam;
   }
   toURL() {
-    const api_version = "v1beta";
+    const api_version = this.model.apiVersion();
     const url = new URL(`${BASE_URL}/${api_version}/models/${this.model}:${this.task}`);
     url.searchParams.append("key", this.apiParam.apikey);
     if (this.stream) {
@@ -676,7 +701,12 @@ async function embeddingProxyHandler(rawReq) {
   log?.warn("request", embedContentRequest);
   let geminiResp = [];
   try {
-    for await (const it of generateContent("embedContent", apiParam, "text-embedding-004", embedContentRequest)) {
+    for await (const it of generateContent(
+      "embedContent",
+      apiParam,
+      new GeminiModel("text-embedding-004"),
+      embedContentRequest
+    )) {
       const data = it.embedding?.values;
       geminiResp = data;
       break;
@@ -745,7 +775,7 @@ app.post("/v1/chat/completions", chatProxyHandler);
 app.post("/v1/embeddings", embeddingProxyHandler);
 app.get("/v1/models", () => Response.json(models()));
 app.get("/v1/models/:model", (c) => Response.json(modelDetail(c.params.model)));
-app.post(":model_version/models/:model_and_action", geminiProxy);
+app.post("/:model_version/models/:model_and_action", geminiProxy);
 app.all("*", () => new Response("Page Not Found", { status: 404 }));
 
 // main_bun.ts
diff --git a/dist/main_cloudflare-workers.mjs b/dist/main_cloudflare-workers.mjs
index 61b58f4..ec1e764 100644
--- a/dist/main_cloudflare-workers.mjs
+++ b/dist/main_cloudflare-workers.mjs
@@ -102,13 +102,7 @@ function openAiMessageToGeminiMessage(messages) {
   return result;
 }
 function genModel(req) {
-  const defaultModel = (m) => {
-    if (m.startsWith("gemini")) {
-      return m;
-    }
-    return "gemini-1.5-flash-latest";
-  };
-  const model = ModelMapping[req.model] ?? defaultModel(req.model);
+  const model = GeminiModel.modelMapping(req.model);
   let functions = req.tools?.filter((it) => it.type === "function")?.map((it) => it.function) ?? [];
   functions = functions.concat((req.functions ?? []).map((it) => ({ strict: null, ...it })));
   const responseMimeType = req.response_format?.type === "json_object" ? "application/json" : "text/plain";
@@ -118,7 +112,10 @@ function genModel(req) {
       maxOutputTokens: req.max_tokens ?? void 0,
       temperature: req.temperature ?? void 0,
       topP: req.top_p ?? void 0,
-      responseMimeType
+      responseMimeType,
+      thinkingConfig: !model.isThinkingModel() ? void 0 : {
+        includeThoughts: true
+      }
     },
     tools: functions.length === 0 ? void 0 : [
       {
@@ -137,6 +134,34 @@ function genModel(req) {
   };
   return [model, generateContentRequest];
 }
+var GeminiModel = class _GeminiModel {
+  static modelMapping(model) {
+    const modelName = ModelMapping[model] ?? _GeminiModel.defaultModel(model);
+    return new _GeminiModel(modelName);
+  }
+  model;
+  constructor(model) {
+    this.model = model;
+  }
+  isThinkingModel() {
+    return this.model.includes("thinking");
+  }
+  apiVersion() {
+    if (this.isThinkingModel()) {
+      return "v1alpha";
+    }
+    return "v1beta";
+  }
+  toString() {
+    return this.model;
+  }
+  static defaultModel(m) {
+    if (m.startsWith("gemini")) {
+      return m;
+    }
+    return "gemini-1.5-flash-latest";
+  }
+};
 var ModelMapping = {
   "gpt-3.5-turbo": "gemini-1.5-flash-8b-latest",
   "gpt-4": "gemini-1.5-pro-latest",
@@ -405,7 +430,7 @@ var RequestUrl = class {
     this.apiParam = apiParam;
   }
   toURL() {
-    const api_version = "v1beta";
+    const api_version = this.model.apiVersion();
     const url = new URL(`${BASE_URL}/${api_version}/models/${this.model}:${this.task}`);
     url.searchParams.append("key", this.apiParam.apikey);
     if (this.stream) {
@@ -676,7 +701,12 @@ async function embeddingProxyHandler(rawReq) {
   log?.warn("request", embedContentRequest);
   let geminiResp = [];
   try {
-    for await (const it of generateContent("embedContent", apiParam, "text-embedding-004", embedContentRequest)) {
+    for await (const it of generateContent(
+      "embedContent",
+      apiParam,
+      new GeminiModel("text-embedding-004"),
+      embedContentRequest
+    )) {
       const data = it.embedding?.values;
       geminiResp = data;
       break;
@@ -745,7 +775,7 @@ app.post("/v1/chat/completions", chatProxyHandler);
 app.post("/v1/embeddings", embeddingProxyHandler);
 app.get("/v1/models", () => Response.json(models()));
 app.get("/v1/models/:model", (c) => Response.json(modelDetail(c.params.model)));
-app.post(":model_version/models/:model_and_action", geminiProxy);
+app.post("/:model_version/models/:model_and_action", geminiProxy);
 app.all("*", () => new Response("Page Not Found", { status: 404 }));
 
 // main_cloudflare-workers.ts
diff --git a/dist/main_deno.mjs b/dist/main_deno.mjs
index a849ac1..198c839 100644
--- a/dist/main_deno.mjs
+++ b/dist/main_deno.mjs
@@ -102,13 +102,7 @@ function openAiMessageToGeminiMessage(messages) {
   return result;
 }
 function genModel(req) {
-  const defaultModel = (m) => {
-    if (m.startsWith("gemini")) {
-      return m;
-    }
-    return "gemini-1.5-flash-latest";
-  };
-  const model = ModelMapping[req.model] ?? defaultModel(req.model);
+  const model = GeminiModel.modelMapping(req.model);
   let functions = req.tools?.filter((it) => it.type === "function")?.map((it) => it.function) ?? [];
   functions = functions.concat((req.functions ?? []).map((it) => ({ strict: null, ...it })));
   const responseMimeType = req.response_format?.type === "json_object" ? "application/json" : "text/plain";
@@ -118,7 +112,10 @@ function genModel(req) {
       maxOutputTokens: req.max_tokens ?? void 0,
       temperature: req.temperature ?? void 0,
       topP: req.top_p ?? void 0,
-      responseMimeType
+      responseMimeType,
+      thinkingConfig: !model.isThinkingModel() ? void 0 : {
+        includeThoughts: true
+      }
     },
     tools: functions.length === 0 ? void 0 : [
       {
@@ -137,6 +134,34 @@ function genModel(req) {
   };
   return [model, generateContentRequest];
 }
+var GeminiModel = class _GeminiModel {
+  static modelMapping(model) {
+    const modelName = ModelMapping[model] ?? _GeminiModel.defaultModel(model);
+    return new _GeminiModel(modelName);
+  }
+  model;
+  constructor(model) {
+    this.model = model;
+  }
+  isThinkingModel() {
+    return this.model.includes("thinking");
+  }
+  apiVersion() {
+    if (this.isThinkingModel()) {
+      return "v1alpha";
+    }
+    return "v1beta";
+  }
+  toString() {
+    return this.model;
+  }
+  static defaultModel(m) {
+    if (m.startsWith("gemini")) {
+      return m;
+    }
+    return "gemini-1.5-flash-latest";
+  }
+};
 var ModelMapping = {
   "gpt-3.5-turbo": "gemini-1.5-flash-8b-latest",
   "gpt-4": "gemini-1.5-pro-latest",
@@ -405,7 +430,7 @@ var RequestUrl = class {
     this.apiParam = apiParam;
   }
   toURL() {
-    const api_version = "v1beta";
+    const api_version = this.model.apiVersion();
     const url = new URL(`${BASE_URL}/${api_version}/models/${this.model}:${this.task}`);
     url.searchParams.append("key", this.apiParam.apikey);
     if (this.stream) {
@@ -676,7 +701,12 @@ async function embeddingProxyHandler(rawReq) {
   log?.warn("request", embedContentRequest);
   let geminiResp = [];
   try {
-    for await (const it of generateContent("embedContent", apiParam, "text-embedding-004", embedContentRequest)) {
+    for await (const it of generateContent(
+      "embedContent",
+      apiParam,
+      new GeminiModel("text-embedding-004"),
+      embedContentRequest
+    )) {
       const data = it.embedding?.values;
       geminiResp = data;
       break;
@@ -745,7 +775,7 @@ app.post("/v1/chat/completions", chatProxyHandler);
 app.post("/v1/embeddings", embeddingProxyHandler);
 app.get("/v1/models", () => Response.json(models()));
 app.get("/v1/models/:model", (c) => Response.json(modelDetail(c.params.model)));
-app.post(":model_version/models/:model_and_action", geminiProxy);
+app.post("/:model_version/models/:model_and_action", geminiProxy);
 app.all("*", () => new Response("Page Not Found", { status: 404 }));
 
 // main_deno.ts
diff --git a/dist/main_node.mjs b/dist/main_node.mjs
index 3de4828..a4b577f 100644
--- a/dist/main_node.mjs
+++ b/dist/main_node.mjs
@@ -547,13 +547,7 @@ function openAiMessageToGeminiMessage(messages) {
   return result;
 }
 function genModel(req) {
-  const defaultModel = (m) => {
-    if (m.startsWith("gemini")) {
-      return m;
-    }
-    return "gemini-1.5-flash-latest";
-  };
-  const model = ModelMapping[req.model] ?? defaultModel(req.model);
+  const model = GeminiModel.modelMapping(req.model);
   let functions = req.tools?.filter((it) => it.type === "function")?.map((it) => it.function) ?? [];
   functions = functions.concat((req.functions ?? []).map((it) => ({ strict: null, ...it })));
   const responseMimeType = req.response_format?.type === "json_object" ? "application/json" : "text/plain";
@@ -563,7 +557,10 @@ function genModel(req) {
       maxOutputTokens: req.max_tokens ?? void 0,
       temperature: req.temperature ?? void 0,
       topP: req.top_p ?? void 0,
-      responseMimeType
+      responseMimeType,
+      thinkingConfig: !model.isThinkingModel() ? void 0 : {
+        includeThoughts: true
+      }
     },
     tools: functions.length === 0 ? void 0 : [
       {
@@ -582,6 +579,34 @@ function genModel(req) {
   };
   return [model, generateContentRequest];
 }
+var GeminiModel = class _GeminiModel {
+  static modelMapping(model) {
+    const modelName = ModelMapping[model] ?? _GeminiModel.defaultModel(model);
+    return new _GeminiModel(modelName);
+  }
+  model;
+  constructor(model) {
+    this.model = model;
+  }
+  isThinkingModel() {
+    return this.model.includes("thinking");
+  }
+  apiVersion() {
+    if (this.isThinkingModel()) {
+      return "v1alpha";
+    }
+    return "v1beta";
+  }
+  toString() {
+    return this.model;
+  }
+  static defaultModel(m) {
+    if (m.startsWith("gemini")) {
+      return m;
+    }
+    return "gemini-1.5-flash-latest";
+  }
+};
 var ModelMapping = {
   "gpt-3.5-turbo": "gemini-1.5-flash-8b-latest",
   "gpt-4": "gemini-1.5-pro-latest",
@@ -850,7 +875,7 @@ var RequestUrl = class {
     this.apiParam = apiParam;
   }
   toURL() {
-    const api_version = "v1beta";
+    const api_version = this.model.apiVersion();
     const url = new URL(`${BASE_URL}/${api_version}/models/${this.model}:${this.task}`);
     url.searchParams.append("key", this.apiParam.apikey);
     if (this.stream) {
@@ -1121,7 +1146,12 @@ async function embeddingProxyHandler(rawReq) {
   log?.warn("request", embedContentRequest);
   let geminiResp = [];
   try {
-    for await (const it of generateContent("embedContent", apiParam, "text-embedding-004", embedContentRequest)) {
+    for await (const it of generateContent(
+      "embedContent",
+      apiParam,
+      new GeminiModel("text-embedding-004"),
+      embedContentRequest
+    )) {
       const data = it.embedding?.values;
       geminiResp = data;
       break;
@@ -1190,7 +1220,7 @@ app.post("/v1/chat/completions", chatProxyHandler);
 app.post("/v1/embeddings", embeddingProxyHandler);
 app.get("/v1/models", () => Response.json(models()));
 app.get("/v1/models/:model", (c) => Response.json(modelDetail(c.params.model)));
-app.post(":model_version/models/:model_and_action", geminiProxy);
+app.post("/:model_version/models/:model_and_action", geminiProxy);
 app.all("*", () => new Response("Page Not Found", { status: 404 }));
 
 // main_node.ts
diff --git a/fly.toml b/fly.toml
index 45ae8b9..f0c9e19 100644
--- a/fly.toml
+++ b/fly.toml
@@ -7,7 +7,7 @@ dockerfile = 'docker/bun.Dockerfile'
 [http_service]
 internal_port = 8000
 force_https = true
-auto_stop_machines = true
+auto_stop_machines = "suspend"
 auto_start_machines = true
 min_machines_running = 0
 processes = ['app']
diff --git a/src/app.ts b/src/app.ts
index 9eb876d..2bf9c48 100644
--- a/src/app.ts
+++ b/src/app.ts
@@ -32,7 +32,7 @@ app.post("/v1/chat/completions", chatProxyHandler)
 app.post("/v1/embeddings", embeddingProxyHandler)
 app.get("/v1/models", () => Response.json(models()))
 app.get("/v1/models/:model", (c) => Response.json(modelDetail(c.params.model)))
-app.post(":model_version/models/:model_and_action", geminiProxy)
+app.post("/:model_version/models/:model_and_action", geminiProxy)
 app.all("*", () => new Response("Page Not Found", { status: 404 }))
 
 export { app }
diff --git a/src/gemini-api-client/gemini-api-client.ts b/src/gemini-api-client/gemini-api-client.ts
index fa532ee..631ff17 100644
--- a/src/gemini-api-client/gemini-api-client.ts
+++ b/src/gemini-api-client/gemini-api-client.ts
@@ -89,8 +89,9 @@ export class RequestUrl {
     this.stream = stream
     this.apiParam = apiParam
   }
+
   toURL(): URL {
-    const api_version: API_VERSION = "v1beta"
+    const api_version = this.model.apiVersion()
     const url = new URL(`${BASE_URL}/${api_version}/models/${this.model}:${this.task}`)
     url.searchParams.append("key", this.apiParam.apikey)
     if (this.stream) {
@@ -102,8 +103,6 @@ export class RequestUrl {
 
 const BASE_URL = "https://generativelanguage.googleapis.com"
 
-type API_VERSION = "v1beta" | "v1"
-
 /**
  * Generates the request options to be passed to the fetch API.
  * @param requestOptions - The user-defined request options.
diff --git a/src/openai/embeddingProxyHandler.ts b/src/openai/embeddingProxyHandler.ts
index b688bb0..c1b585e 100644
--- a/src/openai/embeddingProxyHandler.ts
+++ b/src/openai/embeddingProxyHandler.ts
@@ -1,7 +1,7 @@
 import { generateContent } from "../gemini-api-client/gemini-api-client.ts"
 import type { EmbedContentRequest } from "../gemini-api-client/types.ts"
 import type { OpenAI } from "../types.ts"
-import { getToken } from "../utils.ts"
+import { GeminiModel, getToken } from "../utils.ts"
 
 export async function embeddingProxyHandler(rawReq: Request): Promise<Response> {
   const req = (await rawReq.json()) as OpenAI.Embeddings.EmbeddingCreateParams
@@ -24,7 +24,12 @@ export async function embeddingProxyHandler(rawReq: Request): Promise<Response>
   let geminiResp: number[] | undefined = []
 
   try {
-    for await (const it of generateContent("embedContent", apiParam, "text-embedding-004", embedContentRequest)) {
+    for await (const it of generateContent(
+      "embedContent",
+      apiParam,
+      new GeminiModel("text-embedding-004"),
+      embedContentRequest,
+    )) {
       const data = it.embedding?.values
       geminiResp = data
       break
diff --git a/src/utils.ts b/src/utils.ts
index 9626150..fe2bc97 100644
--- a/src/utils.ts
+++ b/src/utils.ts
@@ -71,14 +71,7 @@ export function openAiMessageToGeminiMessage(messages: OpenAI.Chat.ChatCompletio
 }
 
 export function genModel(req: OpenAI.Chat.ChatCompletionCreateParams): [GeminiModel, GenerateContentRequest] {
-  const defaultModel = (m: string): GeminiModel => {
-    if (m.startsWith("gemini")) {
-      return m as GeminiModel
-    }
-    return "gemini-1.5-flash-latest"
-  }
-
-  const model: GeminiModel = ModelMapping[req.model] ?? defaultModel(req.model)
+  const model: GeminiModel = GeminiModel.modelMapping(req.model)
 
   let functions: OpenAI.Chat.FunctionObject[] =
     req.tools?.filter((it) => it.type === "function")?.map((it) => it.function) ?? []
@@ -94,6 +87,11 @@ export function genModel(req: OpenAI.Chat.ChatCompletionCreateParams): [GeminiMo
       temperature: req.temperature ?? undefined,
       topP: req.top_p ?? undefined,
       responseMimeType: responseMimeType,
+      thinkingConfig: !model.isThinkingModel()
+        ? undefined
+        : {
+            includeThoughts: true,
+          },
     },
     tools:
       functions.length === 0
@@ -124,7 +122,42 @@ export type KnownGeminiModel =
   | "gemini-2.0-flash-exp"
   | "text-embedding-004"
 
-export type GeminiModel = `gemini${string}` | "text-embedding-004"
+export type API_VERSION = "v1beta" | "v1" | "v1alpha"
+
+export class GeminiModel {
+  static modelMapping(model: string): GeminiModel {
+    const modelName: GeminiModelName | KnownGeminiModel = ModelMapping[model] ?? GeminiModel.defaultModel(model)
+    return new GeminiModel(modelName)
+  }
+  public readonly model: GeminiModelName
+  constructor(model: GeminiModelName) {
+    this.model = model
+  }
+
+  isThinkingModel(): boolean {
+    return this.model.includes("thinking")
+  }
+
+  apiVersion(): API_VERSION {
+    if (this.isThinkingModel()) {
+      return "v1alpha"
+    }
+    return "v1beta"
+  }
+
+  toString(): string {
+    return this.model
+  }
+
+  private static defaultModel(m: string): GeminiModelName {
+    if (m.startsWith("gemini")) {
+      return m as GeminiModelName
+    }
+    return "gemini-1.5-flash-latest"
+  }
+}
+
+export type GeminiModelName = `gemini${string}` | "text-embedding-004"
 
 export const ModelMapping: Readonly<Record<string, KnownGeminiModel>> = {
   "gpt-3.5-turbo": "gemini-1.5-flash-8b-latest",