diff --git a/src/main/main.ts b/src/main/main.ts
index fbc2c53f..543bbbb0 100644
--- a/src/main/main.ts
+++ b/src/main/main.ts
@@ -161,8 +161,27 @@ function getNativeBinaryPath(name: string): string {
 }
 
 const WHISPERCPP_FRAMEWORK_VERSION = 'v1.8.3';
-const WHISPERCPP_MODEL_NAME = 'base';
-const WHISPERCPP_MODEL_URL = `https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-${WHISPERCPP_MODEL_NAME}.bin`;
+const WHISPERCPP_DEFAULT_MODEL_SIZE = 'base';
+const WHISPERCPP_MODEL_SIZES = ['tiny', 'base', 'small', 'medium', 'large'] as const;
+let whisperCppModelSizeCache: string | null = null;
+
+function getWhisperCppModelSize(): string {
+  if (whisperCppModelSizeCache) return whisperCppModelSizeCache;
+  try {
+    const s = loadSettings();
+    const size = (s as any).ai?.whisperCppModelSize;
+    if (size && WHISPERCPP_MODEL_SIZES.includes(size)) {
+      whisperCppModelSizeCache = size;
+      return size;
+    }
+  } catch {}
+  whisperCppModelSizeCache = WHISPERCPP_DEFAULT_MODEL_SIZE;
+  return WHISPERCPP_DEFAULT_MODEL_SIZE;
+}
+
+function getWhisperCppModelUrl(size: string): string {
+  return `https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-${size}.bin`;
+}
 
 let whisperCppModelEnsurePromise: Promise<string> | null = null;
 type WhisperCppModelStatus = {
@@ -799,7 +818,8 @@ function getWhisperCppTranscriberBinaryPath(): string {
 }
 
 function getWhisperCppModelPath(): string {
-  return path.join(app.getPath('userData'), 'whispercpp', 'models', `ggml-${WHISPERCPP_MODEL_NAME}.bin`);
+  const size = getWhisperCppModelSize();
+  return path.join(app.getPath('userData'), 'whispercpp', 'models', `ggml-${size}.bin`);
 }
 
 function getWhisperCppModelStatus(): WhisperCppModelStatus {
@@ -809,7 +829,7 @@ function getWhisperCppModelStatus(): WhisperCppModelStatus {
       const stats = fs.statSync(modelPath);
       whisperCppModelStatus = {
         state: 'downloaded',
-        modelName: WHISPERCPP_MODEL_NAME,
+        modelName: getWhisperCppModelSize(),
         path: modelPath,
         bytesDownloaded: Math.max(0, Number(stats.size) || 0),
         totalBytes: Math.max(0, Number(stats.size) || 0),
@@ -821,7 +841,7 @@ function getWhisperCppModelStatus(): WhisperCppModelStatus {
   if (whisperCppModelStatus?.state === 'downloading') {
     return {
       ...whisperCppModelStatus,
-      modelName: WHISPERCPP_MODEL_NAME,
+      modelName: getWhisperCppModelSize(),
       path: modelPath,
     };
   }
@@ -829,14 +849,14 @@ function getWhisperCppModelStatus(): WhisperCppModelStatus {
   if (whisperCppModelStatus?.state === 'error') {
     return {
       ...whisperCppModelStatus,
-      modelName: WHISPERCPP_MODEL_NAME,
+      modelName: getWhisperCppModelSize(),
       path: modelPath,
     };
   }
 
   whisperCppModelStatus = {
     state: 'not-downloaded',
-    modelName: WHISPERCPP_MODEL_NAME,
+    modelName: getWhisperCppModelSize(),
     path: modelPath,
     bytesDownloaded: 0,
     totalBytes: null,
@@ -955,12 +975,13 @@ async function downloadFileWithRedirects(
 }
 
 async function ensureWhisperCppModelDownloaded(): Promise<string> {
+  const size = getWhisperCppModelSize();
   const modelPath = getWhisperCppModelPath();
   try {
     if (fs.existsSync(modelPath)) {
       whisperCppModelStatus = {
         state: 'downloaded',
-        modelName: WHISPERCPP_MODEL_NAME,
+        modelName: size,
         path: modelPath,
         bytesDownloaded: Math.max(0, Number(fs.statSync(modelPath).size) || 0),
         totalBytes: Math.max(0, Number(fs.statSync(modelPath).size) || 0),
@@ -979,18 +1000,18 @@ async function ensureWhisperCppModelDownloaded(): Promise<string> {
     fs.mkdirSync(modelDir, { recursive: true });
     whisperCppModelStatus = {
       state: 'downloading',
-      modelName: WHISPERCPP_MODEL_NAME,
+      modelName: size,
       path: modelPath,
       bytesDownloaded: 0,
       totalBytes: null,
     };
 
     try {
-      console.log(`[Whisper][whisper.cpp] Downloading ${WHISPERCPP_MODEL_NAME} model`);
-      await downloadFileWithRedirects(WHISPERCPP_MODEL_URL, tempPath, 5, (bytesDownloaded, totalBytes) => {
+      console.log(`[Whisper][whisper.cpp] Downloading ${size} model`);
+      await downloadFileWithRedirects(getWhisperCppModelUrl(size), tempPath, 5, (bytesDownloaded, totalBytes) => {
         whisperCppModelStatus = {
           state: 'downloading',
-          modelName: WHISPERCPP_MODEL_NAME,
+          modelName: size,
           path: modelPath,
           bytesDownloaded,
           totalBytes,
@@ -1000,7 +1021,7 @@ async function ensureWhisperCppModelDownloaded(): Promise<string> {
       const finalSize = Math.max(0, Number(fs.statSync(modelPath).size) || 0);
       whisperCppModelStatus = {
         state: 'downloaded',
-        modelName: WHISPERCPP_MODEL_NAME,
+        modelName: size,
         path: modelPath,
         bytesDownloaded: finalSize,
         totalBytes: finalSize,
@@ -1011,7 +1032,7 @@ async function ensureWhisperCppModelDownloaded(): Promise<string> {
       try { fs.unlinkSync(tempPath); } catch {}
       whisperCppModelStatus = {
         state: 'error',
-        modelName: WHISPERCPP_MODEL_NAME,
+        modelName: size,
         path: modelPath,
         bytesDownloaded: 0,
         totalBytes: null,
@@ -11095,6 +11116,10 @@ app.whenReady().then(async () => {
       if (patch.openAtLogin !== undefined) {
         applyOpenAtLogin(Boolean(patch.openAtLogin));
       }
+      // Invalidate cached model size when it changes
+      if (patch.ai?.whisperCppModelSize !== undefined) {
+        whisperCppModelSizeCache = null;
+      }
       // When onboarding completes: hide dock, then start services that were
       // deferred to avoid triggering permission dialogs during onboarding.
       if (patch.hasSeenOnboarding === true) {
@@ -13579,7 +13604,7 @@ if let tiff = image?.tiffRepresentation {
 
       // Parse speechToTextModel to a concrete provider/model pair.
       let provider: 'parakeet' | 'qwen3' | 'whispercpp' | 'openai' | 'elevenlabs' = 'whispercpp';
-      let model = `ggml-${WHISPERCPP_MODEL_NAME}`;
+      let model = `ggml-${getWhisperCppModelSize()}`;
       const sttModel = s.ai.speechToTextModel || '';
       if (sttModel === 'parakeet') {
         provider = 'parakeet';
@@ -13589,7 +13614,7 @@ if let tiff = image?.tiffRepresentation {
         model = 'qwen3-asr-0.6b';
       } else if (!sttModel || sttModel === 'default' || sttModel === 'whispercpp') {
         provider = 'whispercpp';
-        model = `ggml-${WHISPERCPP_MODEL_NAME}`;
+        model = `ggml-${getWhisperCppModelSize()}`;
       } else if (sttModel === 'native') {
         // Renderer should not call cloud transcription in native mode.
         // Return empty transcript instead of surfacing an IPC error.
diff --git a/src/main/settings-store.ts b/src/main/settings-store.ts
index 709967cc..d91f8ac9 100644
--- a/src/main/settings-store.ts
+++ b/src/main/settings-store.ts
@@ -30,6 +30,7 @@ export interface AISettings {
   enabled: boolean;
   llmEnabled: boolean;
   whisperEnabled: boolean;
+  whisperCppModelSize: string;
   readEnabled: boolean;
   openaiCompatibleBaseUrl: string;
   openaiCompatibleApiKey: string;
@@ -125,6 +126,7 @@ const DEFAULT_AI_SETTINGS: AISettings = {
   enabled: true,
   llmEnabled: true,
   whisperEnabled: true,
+  whisperCppModelSize: 'base',
   readEnabled: true,
   openaiCompatibleBaseUrl: '',
   openaiCompatibleApiKey: '',
diff --git a/src/renderer/src/i18n/locales/en.json b/src/renderer/src/i18n/locales/en.json
index 813b6c58..2ce6e928 100644
--- a/src/renderer/src/i18n/locales/en.json
+++ b/src/renderer/src/i18n/locales/en.json
@@ -343,6 +343,16 @@
         "elevenlabsWarning": "ElevenLabs STT selected. {action} ElevenLabs API key in API Keys.",
         "elevenlabsReady": "ElevenLabs STT selected. Cloud transcription will use your ElevenLabs key.",
         "recognitionLanguage": "Recognition Language",
+        "modelSize": {
+          "label": "Model Size",
+          "tiny": "Tiny (~39 MB) — Fastest, lower accuracy",
+          "base": "Base (~140 MB) — Good balance (default)",
+          "small": "Small (~466 MB) — Better accuracy",
+          "medium": "Medium (~1.5 GB) — High accuracy",
+          "large": "Large (~2.9 GB) — Best accuracy, slowest",
+          "notDownloaded": "This model is not downloaded yet. Download it below before using dictation.",
+          "hint": "Larger models are more accurate but use more memory and take longer to transcribe. Changing the model size requires re-downloading."
+        },
         "providerInfo": {
           "parakeet": "Offline on-device transcription via Parakeet TDT v3. Runs on Apple Neural Engine. Requires model warmup on first use. Download the model below before using dictation.",
           "qwen3": "Offline on-device transcription via Qwen3 ASR. Supports 30+ languages, requires macOS 15+, and warms up on first use.",
diff --git a/src/renderer/src/settings/AITab.tsx b/src/renderer/src/settings/AITab.tsx
index b4f39294..e8045ef5 100644
--- a/src/renderer/src/settings/AITab.tsx
+++ b/src/renderer/src/settings/AITab.tsx
@@ -1299,6 +1299,34 @@ const AITab: React.FC = () => {
                 </div>
               )}
 
+              {whisperModelValue === 'whispercpp' && (
+                <div>
+                  <label className="text-[0.75rem] text-[var(--text-muted)] mb-1 block">{t('settings.ai.whisper.modelSize.label')}</label>
+                  <select
+                    value={ai.whisperCppModelSize || 'base'}
+                    onChange={(e) => {
+                      updateAI({ whisperCppModelSize: e.target.value });
+                      void refreshWhisperCppModelStatus();
+                    }}
+                    className={`w-full bg-[var(--ui-segment-bg)] border rounded-md px-2.5 py-2 text-sm text-[var(--text-secondary)] focus:outline-none focus:border-blue-500/50 ${
+                      whisperCppModelStatus && whisperCppModelStatus.state !== 'downloaded' && whisperCppModelStatus.state !== 'downloading'
+                        ? 'border-red-500/60'
+                        : 'border-[var(--ui-divider)]'
+                    }`}
+                  >
+                    <option value="tiny">{t('settings.ai.whisper.modelSize.tiny')}</option>
+                    <option value="base">{t('settings.ai.whisper.modelSize.base')}</option>
+                    <option value="small">{t('settings.ai.whisper.modelSize.small')}</option>
+                    <option value="medium">{t('settings.ai.whisper.modelSize.medium')}</option>
+                    <option value="large">{t('settings.ai.whisper.modelSize.large')}</option>
+                  </select>
+                  {whisperCppModelStatus && whisperCppModelStatus.state !== 'downloaded' && whisperCppModelStatus.state !== 'downloading' && (
+                    <p className="text-[0.6875rem] text-red-400 mt-1">{t('settings.ai.whisper.modelSize.notDownloaded')}</p>
+                  )}
+                  <p className="text-[0.75rem] text-[var(--text-muted)] mt-1">{t('settings.ai.whisper.modelSize.hint')}</p>
+                </div>
+              )}
+
               {whisperModelValue === 'whispercpp' && (
                 <div>
                   <label className="text-[0.75rem] text-[var(--text-muted)] mb-1 block">{t('settings.ai.whisper.recognitionLanguage')}</label>
diff --git a/src/renderer/types/electron.d.ts b/src/renderer/types/electron.d.ts
index 6d12fa39..cbd5986b 100644
--- a/src/renderer/types/electron.d.ts
+++ b/src/renderer/types/electron.d.ts
@@ -175,6 +175,7 @@ export interface AISettings {
   enabled: boolean;
   llmEnabled: boolean;
   whisperEnabled: boolean;
+  whisperCppModelSize: string;
   readEnabled: boolean;
   openaiCompatibleBaseUrl: string;
   openaiCompatibleApiKey: string;