diff --git a/packages/ai/src/protocols/gemini.ts b/packages/ai/src/protocols/gemini.ts
index 82a69b059e..82759f0eec 100644
--- a/packages/ai/src/protocols/gemini.ts
+++ b/packages/ai/src/protocols/gemini.ts
@@ -93,8 +93,10 @@ const GeminiToolConfig = Schema.Struct({
 
 const GeminiThinkingConfig = Schema.Struct({
   thinkingBudget: Schema.optional(Schema.Number),
+  thinkingLevel: Schema.optional(Schema.Literals(["minimal", "low", "medium", "high"])),
   includeThoughts: Schema.optional(Schema.Boolean),
 })
+type GeminiThinkingLevel = NonNullable<Schema.Schema.Type<typeof GeminiThinkingConfig>["thinkingLevel"]>
 
 const GeminiGenerationConfig = Schema.Struct({
   maxOutputTokens: Schema.optional(Schema.Number),
@@ -289,11 +291,16 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
 
 const geminiOptions = (request: LLMRequest) => request.providerOptions?.gemini
 
+const isThinkingLevel = (value: unknown): value is GeminiThinkingLevel =>
+  value === "minimal" || value === "low" || value === "medium" || value === "high"
+
 const thinkingConfig = (request: LLMRequest) => {
   const value = geminiOptions(request)?.thinkingConfig
   if (!ProviderShared.isRecord(value)) return undefined
+  const thinkingLevel = value.thinkingLevel
   const result = {
     thinkingBudget: typeof value.thinkingBudget === "number" ? value.thinkingBudget : undefined,
+    thinkingLevel: isThinkingLevel(thinkingLevel) ? thinkingLevel : undefined,
     includeThoughts: typeof value.includeThoughts === "boolean" ? value.includeThoughts : undefined,
   }
   return Object.values(result).some((item) => item !== undefined) ? result : undefined
diff --git a/packages/ai/test/provider/gemini.test.ts b/packages/ai/test/provider/gemini.test.ts
index 1dc253c0ea..291f86c8f9 100644
--- a/packages/ai/test/provider/gemini.test.ts
+++ b/packages/ai/test/provider/gemini.test.ts
@@ -36,6 +36,20 @@ describe("Gemini route", () => {
     }),
   )
 
+  it.effect("prepares thinking level", () =>
+    Effect.gen(function* () {
+      const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(
+        LLM.request({
+          model,
+          prompt: "Say hello.",
+          providerOptions: { gemini: { thinkingConfig: { thinkingLevel: "minimal" } } },
+        }),
+      )
+
+      expect(prepared.body.generationConfig).toEqual({ thinkingConfig: { thinkingLevel: "minimal" } })
+    }),
+  )
+
   it.effect("lowers chronological system updates to wrapped user text in order", () =>
     Effect.gen(function* () {
       const prepared = yield* LLMClient.prepare<Gemini.GeminiBody>(