diff --git a/.env.example b/.env.example
index 0493550..0fb0128 100644
--- a/.env.example
+++ b/.env.example
@@ -1,5 +1,2 @@
 DISCORD_TOKEN=
 GEMINI_API_KEY=
-GEMINI_MODEL=
-# Opcional: se usa si GEMINI_MODEL sigue saturado después de los reintentos
-GEMINI_FALLBACK_MODEL=
diff --git a/src/gemini.ts b/src/gemini.ts
index 0aabf07..57eb50f 100644
--- a/src/gemini.ts
+++ b/src/gemini.ts
@@ -11,8 +11,7 @@ import type { AudioTrack } from './recorder.js';
 import { buildPrompt } from './prompt.js';
 
 const ai = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY });
-export const MODEL = process.env.GEMINI_MODEL!;
-export const FALLBACK_MODEL = process.env.GEMINI_FALLBACK_MODEL;
+export const MODEL = 'gemini-3.5-flash-lite';
 
 const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
 
@@ -26,28 +25,22 @@ async function upload(track: AudioTrack): Promise<File> {
   return file;
 }
 
-// 429/500/503 suelen ser picos de demanda: reintenta y, si hay GEMINI_FALLBACK_MODEL, prueba con ese
-const RETRY_DELAYS = [5_000, 15_000, 30_000];
+// Los errores temporales de Gemini pueden resolverse con reintentos del mismo modelo.
+const RETRY_DELAYS = [2_000, 4_000, 8_000, 16_000, 32_000];
 
 async function generateWithRetry(contents: Content) {
-  const models = [MODEL, FALLBACK_MODEL].filter((m): m is string => !!m);
-  let lastErr: unknown;
-  for (const model of models) {
-    for (let attempt = 0; attempt <= RETRY_DELAYS.length; attempt++) {
-      try {
-        return await ai.models.generateContent({ model, contents });
-      } catch (err) {
-        lastErr = err;
-        const status = err instanceof ApiError ? err.status : 0;
-        if (![429, 500, 503].includes(status)) throw err;
-        const delay = RETRY_DELAYS[attempt];
-        if (delay === undefined) break;
-        console.warn(`${model} respondió ${status}, reintento en ${delay / 1000}s`);
-        await sleep(delay);
-      }
+  for (let attempt = 0; ; attempt++) {
+    try {
+      return await ai.models.generateContent({ model: MODEL, contents });
+    } catch (err) {
+      const status = err instanceof ApiError ? err.status : 0;
+      if (![429, 500, 503].includes(status)) throw err;
+      const delay = RETRY_DELAYS[attempt];
+      if (delay === undefined) throw err;
+      console.warn(`${MODEL} respondió ${status}; reintento ${attempt + 1}/${RETRY_DELAYS.length} en ${delay / 1000}s`);
+      await sleep(delay);
     }
   }
-  throw lastErr;
 }
 
 export async function generateNotes(tracks: AudioTrack[], startedAt: Date): Promise<string> {
diff --git a/src/index.ts b/src/index.ts
index 68aa453..ff88d0b 100644
--- a/src/index.ts
+++ b/src/index.ts
@@ -9,9 +9,9 @@ import {
 } from 'discord.js';
 import { VoiceConnectionStatus, entersState } from '@discordjs/voice';
 import { Recording } from './recorder.js';
-import { FALLBACK_MODEL, MODEL, generateNotes } from './gemini.js';
+import { MODEL, generateNotes } from './gemini.js';
 
-for (const key of ['DISCORD_TOKEN', 'GEMINI_API_KEY', 'GEMINI_MODEL']) {
+for (const key of ['DISCORD_TOKEN', 'GEMINI_API_KEY']) {
   if (!process.env[key]) throw new Error(`Falta la variable de entorno ${key}`);
 }
 
@@ -62,7 +62,7 @@ async function autoFinish(guildId: string, reason: string) {
 
 client.once(Events.ClientReady, async (c) => {
   await c.application.commands.set(commands.map((cmd) => cmd.toJSON()));
-  console.log(`ready as ${c.user.tag} | modelo: ${MODEL} | respaldo: ${FALLBACK_MODEL ?? '(ninguno)'}`);
+  console.log(`ready as ${c.user.tag} | modelo único: ${MODEL} | reintentos activos`);
 });
 
 client.on(Events.InteractionCreate, async (interaction) => {