diff --git a/mobile/app/voice-settings.tsx b/mobile/app/voice-settings.tsx
index cfee1854f..a6c725c0e 100644
--- a/mobile/app/voice-settings.tsx
+++ b/mobile/app/voice-settings.tsx
@@ -19,6 +19,7 @@ import type { RpcClient } from '../src/transport/rpc-client'
 import { BottomDrawer } from '../src/components/BottomDrawer'
 import { VoiceModelList } from '../src/components/VoiceModelList'
 import {
+  deleteDictationModel,
   downloadDictationModel,
   fetchDictationSetup,
   isModelInFlight,
@@ -34,6 +35,8 @@ const DICTATION_MODES = [
   { value: 'hold', label: 'Hold' }
 ] as const
 
+type ModelBusyAction = { modelId: string; type: 'download' | 'select' | 'delete' }
+
 export default function VoiceSettingsScreen(): React.JSX.Element {
   const router = useRouter()
   const insets = useSafeAreaInsets()
@@ -53,7 +56,7 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
   const [setup, setSetup] = useState<MobileSpeechSetup | null>(null)
   const [loading, setLoading] = useState(false)
   const [error, setError] = useState<string | null>(null)
-  const [busyModelId, setBusyModelId] = useState<string | null>(null)
+  const [busyAction, setBusyAction] = useState<ModelBusyAction | null>(null)
   const [modelDrawerOpen, setModelDrawerOpen] = useState(false)
   const pollRef = useRef<ReturnType<typeof setInterval> | null>(null)
 
@@ -134,7 +137,7 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
       if (!client) {
         return
       }
-      setBusyModelId(model.id)
+      setBusyAction({ modelId: model.id, type: 'select' })
       setError(null)
       try {
         setSetup(await setDictationConfig(client, { enabled: true, modelId: model.id }))
@@ -142,7 +145,7 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
       } catch (err) {
         setError(err instanceof Error ? err.message : 'Could not select model')
       } finally {
-        setBusyModelId(null)
+        setBusyAction(null)
       }
     },
     [client]
@@ -153,7 +156,7 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
       if (!client) {
         return
       }
-      setBusyModelId(model.id)
+      setBusyAction({ modelId: model.id, type: 'download' })
       setError(null)
       try {
         await downloadDictationModel(client, model.id)
@@ -161,12 +164,34 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
       } catch (err) {
         setError(err instanceof Error ? err.message : 'Download failed')
       } finally {
-        setBusyModelId(null)
+        setBusyAction(null)
       }
     },
     [client, refresh]
   )
 
+  const handleDelete = useCallback(
+    async (model: MobileSpeechModel) => {
+      if (!client) {
+        return
+      }
+      const deletedSelectedModel = setup?.selectedModelId === model.id
+      setBusyAction({ modelId: model.id, type: 'delete' })
+      setError(null)
+      try {
+        setSetup(await deleteDictationModel(client, model.id))
+        if (deletedSelectedModel) {
+          setModelDrawerOpen(false)
+        }
+      } catch (err) {
+        setError(err instanceof Error ? err.message : 'Delete failed')
+      } finally {
+        setBusyAction(null)
+      }
+    },
+    [client, setup?.selectedModelId]
+  )
+
   const enabled = setup?.enabled ?? false
   const selectedModel = setup?.models.find((m) => m.id === setup.selectedModelId)
   const selectedModelLabel = selectedModel?.label ?? 'None selected'
@@ -276,9 +301,10 @@ export default function VoiceSettingsScreen(): React.JSX.Element {
           <VoiceModelList
             setup={setup}
             disabled={false}
-            busyModelId={busyModelId}
+            busyAction={busyAction}
             onUseModel={(m) => void handleUseModel(m)}
             onDownload={(m) => void handleDownload(m)}
+            onDelete={(m) => void handleDelete(m)}
           />
         ) : null}
       </BottomDrawer>
diff --git a/mobile/src/components/VoiceModelList.tsx b/mobile/src/components/VoiceModelList.tsx
index 4f74d18b5..4151f2591 100644
--- a/mobile/src/components/VoiceModelList.tsx
+++ b/mobile/src/components/VoiceModelList.tsx
@@ -1,5 +1,5 @@
 import { ActivityIndicator, Pressable, StyleSheet, Text, View } from 'react-native'
-import { Check, Download } from 'lucide-react-native'
+import { Check, Download, Trash2 } from 'lucide-react-native'
 import { colors, radii, spacing, typography } from '../theme/mobile-theme'
 import {
   isModelInFlight,
@@ -11,9 +11,10 @@ type Props = {
   setup: MobileSpeechSetup
   // Disabled mirrors desktop: the model list greys out when dictation is off.
   disabled: boolean
-  busyModelId: string | null
+  busyAction: { modelId: string; type: 'download' | 'select' | 'delete' } | null
   onUseModel: (model: MobileSpeechModel) => void
   onDownload: (model: MobileSpeechModel) => void
+  onDelete: (model: MobileSpeechModel) => void
 }
 
 function formatSize(bytes: number | null): string {
@@ -38,27 +39,34 @@ function modelMeta(model: MobileSpeechModel): string {
 }
 
 // Renders the speech-model rows shared between the setup sheet and the Voice
-// settings page: size/progress, recommended badge, selected check, download.
+// settings page: size/progress, recommended badge, selected check, download, delete.
 export function VoiceModelList({
   setup,
   disabled,
-  busyModelId,
+  busyAction,
   onUseModel,
-  onDownload
+  onDownload,
+  onDelete
 }: Props): React.JSX.Element {
   return (
     <View style={disabled ? styles.disabled : undefined} pointerEvents={disabled ? 'none' : 'auto'}>
       {setup.models.map((model, idx) => {
+        const anyBusy = busyAction !== null
         const isSelected = model.id === setup.selectedModelId
         const inFlight = isModelInFlight(model)
-        const rowBusy = busyModelId === model.id
+        const rowBusy = busyAction?.modelId === model.id
+        const selectBusy = rowBusy && busyAction?.type === 'select'
+        const downloadBusy = rowBusy && busyAction?.type === 'download'
+        const deleteBusy = rowBusy && busyAction?.type === 'delete'
         return (
           <View key={model.id}>
             {idx > 0 && <View style={styles.separator} />}
             <View style={styles.modelRow}>
               <View style={styles.modelInfo}>
                 <View style={styles.modelTitleRow}>
-                  <Text style={styles.modelLabel}>{model.label}</Text>
+                  <Text style={styles.modelLabel} numberOfLines={1}>
+                    {model.label}
+                  </Text>
                   {model.recommended ? <Text style={styles.recommended}>Recommended</Text> : null}
                 </View>
                 <Text style={styles.modelMeta}>{modelMeta(model)}</Text>
@@ -68,30 +76,51 @@ export function VoiceModelList({
                   {model.status === 'ready' ? 'API key set' : 'Set up on desktop'}
                 </Text>
               ) : model.status === 'ready' ? (
-                isSelected ? (
-                  <View style={styles.selectedTag}>
-                    <Check size={14} color={colors.statusGreen} strokeWidth={2.4} />
-                    <Text style={styles.selectedText}>In use</Text>
-                  </View>
-                ) : (
+                <View style={styles.readyActions}>
+                  {isSelected ? (
+                    <View style={styles.selectedTag}>
+                      <Check size={14} color={colors.statusGreen} strokeWidth={2.4} />
+                      <Text style={styles.selectedText}>In use</Text>
+                    </View>
+                  ) : (
+                    <Pressable
+                      style={({ pressed }) => [
+                        styles.actionButton,
+                        pressed && styles.actionPressed
+                      ]}
+                      disabled={anyBusy}
+                      onPress={() => onUseModel(model)}
+                    >
+                      {selectBusy ? (
+                        <ActivityIndicator size="small" color={colors.textSecondary} />
+                      ) : (
+                        <Text style={styles.actionText}>Use</Text>
+                      )}
+                    </Pressable>
+                  )}
                   <Pressable
-                    style={({ pressed }) => [styles.actionButton, pressed && styles.actionPressed]}
-                    disabled={rowBusy}
-                    onPress={() => onUseModel(model)}
+                    style={({ pressed }) => [styles.iconButton, pressed && styles.actionPressed]}
+                    disabled={anyBusy}
+                    onPress={() => onDelete(model)}
+                    accessibilityLabel={'Delete ' + model.label}
                   >
-                    <Text style={styles.actionText}>Use</Text>
+                    {deleteBusy ? (
+                      <ActivityIndicator size="small" color={colors.statusRed} />
+                    ) : (
+                      <Trash2 size={18} color={colors.statusRed} strokeWidth={2.2} />
+                    )}
                   </Pressable>
-                )
+                </View>
               ) : inFlight ? (
                 <ActivityIndicator size="small" color={colors.textSecondary} />
               ) : (
                 <Pressable
                   style={({ pressed }) => [styles.iconButton, pressed && styles.actionPressed]}
-                  disabled={rowBusy}
+                  disabled={anyBusy}
                   onPress={() => onDownload(model)}
                   accessibilityLabel={'Download ' + model.label}
                 >
-                  {rowBusy ? (
+                  {downloadBusy ? (
                     <ActivityIndicator size="small" color={colors.textSecondary} />
                   ) : (
                     <Download size={18} color={colors.textSecondary} strokeWidth={2.2} />
@@ -118,7 +147,12 @@ const styles = StyleSheet.create({
   },
   modelInfo: { flex: 1, minWidth: 0 },
   modelTitleRow: { flexDirection: 'row', alignItems: 'center', gap: spacing.sm },
-  modelLabel: { color: colors.textPrimary, fontSize: typography.bodySize, fontWeight: '500' },
+  modelLabel: {
+    color: colors.textPrimary,
+    fontSize: typography.bodySize,
+    fontWeight: '500',
+    flexShrink: 1
+  },
   recommended: { color: colors.statusGreen, fontSize: 10, fontWeight: '700' },
   modelMeta: { color: colors.textMuted, fontSize: typography.metaSize, marginTop: 2 },
   modelStateText: { color: colors.textMuted, fontSize: typography.metaSize },
@@ -141,6 +175,7 @@ const styles = StyleSheet.create({
     justifyContent: 'center',
     backgroundColor: colors.bgRaised
   },
+  readyActions: { flexDirection: 'row', alignItems: 'center', gap: spacing.xs },
   selectedTag: { flexDirection: 'row', alignItems: 'center', gap: 4 },
   selectedText: { color: colors.statusGreen, fontSize: typography.metaSize, fontWeight: '600' },
   separator: {
diff --git a/mobile/src/dictation/mobile-dictation-setup.test.ts b/mobile/src/dictation/mobile-dictation-setup.test.ts
index 375f989f1..d72854afc 100644
--- a/mobile/src/dictation/mobile-dictation-setup.test.ts
+++ b/mobile/src/dictation/mobile-dictation-setup.test.ts
@@ -2,6 +2,7 @@ import { describe, expect, it, vi } from 'vitest'
 import type { RpcClient } from '../transport/rpc-client'
 import type { RpcFailure, RpcResponse, RpcSuccess } from '../transport/types'
 import {
+  deleteDictationModel,
   downloadDictationModel,
   fetchDictationSetup,
   isDictationReady,
@@ -72,6 +73,13 @@ describe('rpc wrappers', () => {
     expect(client.calls[0]).toEqual({ method: 'speech.models.download', params: { modelId: 'm1' } })
   })
 
+  it('deletes a model and returns refreshed setup', async () => {
+    const setup: MobileSpeechSetup = { enabled: true, selectedModelId: '', models: [] }
+    const client = clientWith([ok(setup)])
+    await expect(deleteDictationModel(client, 'm1')).resolves.toEqual(setup)
+    expect(client.calls[0]).toEqual({ method: 'speech.models.delete', params: { modelId: 'm1' } })
+  })
+
   it('sets config', async () => {
     const setup: MobileSpeechSetup = { enabled: true, selectedModelId: 'm1', models: [] }
     const client = clientWith([ok(setup)])
diff --git a/mobile/src/dictation/mobile-dictation-setup.ts b/mobile/src/dictation/mobile-dictation-setup.ts
index 489465710..ad4d4b968 100644
--- a/mobile/src/dictation/mobile-dictation-setup.ts
+++ b/mobile/src/dictation/mobile-dictation-setup.ts
@@ -51,6 +51,17 @@ export async function downloadDictationModel(
   }
 }
 
+export async function deleteDictationModel(
+  client: Pick<RpcClient, 'sendRequest'>,
+  modelId: string
+): Promise<MobileSpeechSetup> {
+  const response = await client.sendRequest('speech.models.delete', { modelId })
+  if (!response.ok) {
+    throw new Error(response.error?.message || 'Failed to delete model')
+  }
+  return (response as RpcSuccess).result as MobileSpeechSetup
+}
+
 export async function setDictationConfig(
   client: Pick<RpcClient, 'sendRequest'>,
   params: { enabled?: boolean; modelId?: string; dictationMode?: 'toggle' | 'hold' }
diff --git a/src/main/ipc/speech.test.ts b/src/main/ipc/speech.test.ts
index 05d6dbb5e..2e84124b8 100644
--- a/src/main/ipc/speech.test.ts
+++ b/src/main/ipc/speech.test.ts
@@ -1,12 +1,18 @@
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 
-const { handleMock, fromWebContentsMock, getSpeechModelManagerMock, getSpeechSttServiceMock } =
-  vi.hoisted(() => ({
-    handleMock: vi.fn(),
-    fromWebContentsMock: vi.fn(),
-    getSpeechModelManagerMock: vi.fn(),
-    getSpeechSttServiceMock: vi.fn()
-  }))
+const {
+  handleMock,
+  fromWebContentsMock,
+  getSpeechModelManagerMock,
+  getSpeechSttServiceMock,
+  deleteLocalSpeechModelMock
+} = vi.hoisted(() => ({
+  handleMock: vi.fn(),
+  fromWebContentsMock: vi.fn(),
+  getSpeechModelManagerMock: vi.fn(),
+  getSpeechSttServiceMock: vi.fn(),
+  deleteLocalSpeechModelMock: vi.fn()
+}))
 
 vi.mock('electron', () => ({
   app: { getPath: vi.fn(() => '/tmp/orca-speech-test') },
@@ -33,6 +39,10 @@ vi.mock('../speech/speech-runtime-service', () => ({
   getSpeechSttService: getSpeechSttServiceMock
 }))
 
+vi.mock('../speech/speech-model-deletion', () => ({
+  deleteLocalSpeechModel: deleteLocalSpeechModelMock
+}))
+
 import { registerSpeechHandlers } from './speech'
 
 type SpeechDownloadHandler = (event: { sender: { id: number } }, modelId: string) => Promise<void>
@@ -51,6 +61,7 @@ describe('registerSpeechHandlers', () => {
     fromWebContentsMock.mockReset()
     getSpeechModelManagerMock.mockReset()
     getSpeechSttServiceMock.mockReset()
+    deleteLocalSpeechModelMock.mockReset()
   })
 
   it('clears the model download progress callback after completion', async () => {
@@ -126,4 +137,23 @@ describe('registerSpeechHandlers', () => {
     expect(clearProgressCallback).toHaveBeenCalledTimes(1)
     expect(window.off).toHaveBeenCalledWith('closed', expect.any(Function))
   })
+
+  it('routes desktop model deletion through the shared deletion helper', async () => {
+    const store = {} as never
+    const manager = { deleteModel: vi.fn() }
+    const sttService = { prepareModelForDeletion: vi.fn() }
+    getSpeechModelManagerMock.mockReturnValue(manager)
+    getSpeechSttServiceMock.mockReturnValue(sttService)
+    deleteLocalSpeechModelMock.mockResolvedValue(undefined)
+    registerSpeechHandlers(store)
+
+    await getHandler('speech:deleteModel')({ sender: { id: 7 } }, 'model-1')
+
+    expect(deleteLocalSpeechModelMock).toHaveBeenCalledWith({
+      store,
+      modelManager: manager,
+      sttService,
+      modelId: 'model-1'
+    })
+  })
 })
diff --git a/src/main/ipc/speech.ts b/src/main/ipc/speech.ts
index 053a0038d..808ec375e 100644
--- a/src/main/ipc/speech.ts
+++ b/src/main/ipc/speech.ts
@@ -2,7 +2,8 @@ import { ipcMain, BrowserWindow, systemPreferences, app } from 'electron'
 import { join } from 'path'
 import { writeFile, unlink } from 'fs/promises'
 import { createHash } from 'crypto'
-import { SPEECH_MODEL_CATALOG, getCatalogModel } from '../speech/model-catalog'
+import { SPEECH_MODEL_CATALOG } from '../speech/model-catalog'
+import { deleteLocalSpeechModel } from '../speech/speech-model-deletion'
 import { getSpeechModelManager, getSpeechSttService } from '../speech/speech-runtime-service'
 import {
   clearOpenAiSpeechApiKey,
@@ -69,10 +70,12 @@ export function registerSpeechHandlers(store: Store): void {
   })
 
   ipcMain.handle('speech:deleteModel', async (_event, modelId: string) => {
-    if (!getCatalogModel(modelId)) {
-      throw new Error(`Unknown model: ${modelId}`)
-    }
-    await getSpeechModelManager(store).deleteModel(modelId)
+    await deleteLocalSpeechModel({
+      store,
+      modelManager: getSpeechModelManager(store),
+      sttService: getSpeechSttService(store),
+      modelId
+    })
   })
 
   const getHotwordsFilePath = (content: string): string => {
diff --git a/src/main/runtime/orca-runtime.ts b/src/main/runtime/orca-runtime.ts
index 1c68dff4b..66b728103 100644
--- a/src/main/runtime/orca-runtime.ts
+++ b/src/main/runtime/orca-runtime.ts
@@ -625,6 +625,10 @@ import type { RateLimitState } from '../../shared/rate-limit-types'
 import type { VoiceSettings } from '../../shared/speech-types'
 import { getSpeechModelManager, getSpeechSttService } from '../speech/speech-runtime-service'
 import { getCatalogModel, isLocalSpeechModel, SPEECH_MODEL_CATALOG } from '../speech/model-catalog'
+import {
+  deleteLocalSpeechModel,
+  getSpeechModelDeletionErrorCode
+} from '../speech/speech-model-deletion'
 import type { CommitMessageAgentEnvironmentResolvers } from '../text-generation/commit-message-agent-environment'
 import { scanNestedRepos } from '../project-groups/nested-repo-discovery'
 import {
@@ -2980,7 +2984,9 @@ export class OrcaRuntimeService {
       .map((group) => {
         const tabOrder = retainedOrder.get(group.id) ?? []
         const keptActive =
-          group.activeTabId && tabOrder.includes(group.activeTabId) && liveTabIds.has(group.activeTabId)
+          group.activeTabId &&
+          tabOrder.includes(group.activeTabId) &&
+          liveTabIds.has(group.activeTabId)
             ? group.activeTabId
             : null
         return {
@@ -3370,8 +3376,7 @@ export class OrcaRuntimeService {
       groupIdByTabId.set(newTabAssignment!.tabId, newTabAssignment!.groupId)
     }
     const activeGroupId =
-      (activeTopLevelId ? groupIdByTabId.get(activeTopLevelId) : undefined) ??
-      existingGroups[0]!.id
+      (activeTopLevelId ? groupIdByTabId.get(activeTopLevelId) : undefined) ?? existingGroups[0]!.id
     const orderByGroup = new Map<string, string[]>(existingGroups.map((group) => [group.id, []]))
     for (const tabId of tabOrder) {
       const groupId = groupIdByTabId.get(tabId) ?? activeGroupId
@@ -4061,14 +4066,12 @@ export class OrcaRuntimeService {
   // Merge the client's pane structure into the persisted tab layout. PTY
   // bindings and active leaf stay host-owned; only ratios/expand/titles change.
   // terminalLayoutsByTabId is keyed by tab id (worktree-independent).
-  private persistHeadlessTerminalPaneLayout(
-    args: {
-      tabId: string
-      root: TerminalPaneLayoutNode | null
-      expandedLeafId: string | null
-      titlesByLeafId?: Record<string, string>
-    }
-  ): void {
+  private persistHeadlessTerminalPaneLayout(args: {
+    tabId: string
+    root: TerminalPaneLayoutNode | null
+    expandedLeafId: string | null
+    titlesByLeafId?: Record<string, string>
+  }): void {
     const session = this.store?.getWorkspaceSession?.()
     if (!session || !this.store?.setWorkspaceSession) {
       return
@@ -4167,11 +4170,7 @@ export class OrcaRuntimeService {
     })
     const active = nextTabs.find((candidate) => candidate.isActive) ?? nextTabs[0] ?? null
     const reorderedTargetActiveTabId =
-      active?.type === 'terminal'
-        ? active.parentTabId
-        : active
-          ? active.id
-          : (tabOrder[0] ?? null)
+      active?.type === 'terminal' ? active.parentTabId : active ? active.id : (tabOrder[0] ?? null)
     // Why: reorder only changes ONE group's order. Preserve every other group so
     // a multi-group split isn't deleted by re-sorting tabs in one of its groups.
     const existingGroups = snapshot.tabGroups ?? []
@@ -4323,9 +4322,7 @@ export class OrcaRuntimeService {
       ...session,
       tabsByWorktree: {
         ...session.tabsByWorktree,
-        [worktreeId]: tabs.map((tab) =>
-          tab.id === tabId ? { ...tab, customTitle: title } : tab
-        )
+        [worktreeId]: tabs.map((tab) => (tab.id === tabId ? { ...tab, customTitle: title } : tab))
       }
     })
   }
@@ -5549,6 +5546,28 @@ export class OrcaRuntimeService {
     return { started: true }
   }
 
+  async deleteMobileSpeechModel(modelId: string): Promise<RuntimeSpeechSetupState> {
+    if (!this.store?.getSettings || !this.store.updateSettings) {
+      throw new Error('voice_dictation_unavailable')
+    }
+    const store = this.store
+    try {
+      // The runtime store is adapted to the minimal speech settings contract used by deletion.
+      await deleteLocalSpeechModel({
+        store: {
+          getSettings: () => store.getSettings(),
+          updateSettings: (updates, options) => store.updateSettings?.(updates, options)
+        },
+        modelManager: getSpeechModelManager(store),
+        sttService: getSpeechSttService(store),
+        modelId
+      })
+    } catch (error) {
+      throw new Error(getSpeechModelDeletionErrorCode(error) ?? 'voice_model_delete_failed')
+    }
+    return this.listMobileSpeechModels()
+  }
+
   // Enables/disables dictation and/or selects the model, merging into the
   // existing voice settings so other voice fields are preserved.
   async configureMobileDictation(params: {
@@ -15118,8 +15137,7 @@ export class OrcaRuntimeService {
       } catch {
         warnings.push({
           code: 'LINEAGE_PARENT_CONTEXT_MISSING',
-          message:
-            'Worktree created, but Orca could not validate the environment parent context.',
+          message: 'Worktree created, but Orca could not validate the environment parent context.',
           details: { envParentWorkspace: input.envParentWorkspace }
         })
       }
diff --git a/src/main/runtime/rpc/methods/speech.test.ts b/src/main/runtime/rpc/methods/speech.test.ts
index 7218836c2..dbc06ab8a 100644
--- a/src/main/runtime/rpc/methods/speech.test.ts
+++ b/src/main/runtime/rpc/methods/speech.test.ts
@@ -102,6 +102,35 @@ describe('speech RPC methods', () => {
     expect(response).toMatchObject({ ok: true, result: { started: true } })
   })
 
+  it('deletes a speech model and returns refreshed setup', async () => {
+    const setup = { enabled: true, selectedModelId: '', dictationMode: 'toggle', models: [] }
+    const runtime = {
+      getRuntimeId: () => 'test-runtime',
+      deleteMobileSpeechModel: vi.fn().mockResolvedValue(setup)
+    } as unknown as OrcaRuntimeService
+    const dispatcher = new RpcDispatcher({ runtime, methods: SPEECH_METHODS })
+
+    const response = await dispatcher.dispatch(
+      makeRequest('speech.models.delete', { modelId: 'parakeet-tdt-0.6b-v3-int8' })
+    )
+
+    expect(runtime.deleteMobileSpeechModel).toHaveBeenCalledWith('parakeet-tdt-0.6b-v3-int8')
+    expect(response).toMatchObject({ ok: true, result: setup })
+  })
+
+  it('rejects invalid speech model delete params', async () => {
+    const runtime = {
+      getRuntimeId: () => 'test-runtime',
+      deleteMobileSpeechModel: vi.fn()
+    } as unknown as OrcaRuntimeService
+    const dispatcher = new RpcDispatcher({ runtime, methods: SPEECH_METHODS })
+
+    const response = await dispatcher.dispatch(makeRequest('speech.models.delete', {}))
+
+    expect(response).toMatchObject({ ok: false })
+    expect(runtime.deleteMobileSpeechModel).not.toHaveBeenCalled()
+  })
+
   it('configures dictation enable + model selection', async () => {
     const runtime = {
       getRuntimeId: () => 'test-runtime',
diff --git a/src/main/runtime/rpc/methods/speech.ts b/src/main/runtime/rpc/methods/speech.ts
index fcbcd5bb0..d686475ab 100644
--- a/src/main/runtime/rpc/methods/speech.ts
+++ b/src/main/runtime/rpc/methods/speech.ts
@@ -38,7 +38,7 @@ const DictationHandle = z.object({
   dictationId: requiredString('Missing dictation ID')
 })
 
-const SpeechModelDownload = z.object({
+const SpeechModelAction = z.object({
   modelId: requiredString('Missing model ID')
 })
 
@@ -56,9 +56,14 @@ export const SPEECH_METHODS: RpcMethod[] = [
   }),
   defineMethod({
     name: 'speech.models.download',
-    params: SpeechModelDownload,
+    params: SpeechModelAction,
     handler: async (params, { runtime }) => runtime.downloadMobileSpeechModel(params.modelId)
   }),
+  defineMethod({
+    name: 'speech.models.delete',
+    params: SpeechModelAction,
+    handler: async (params, { runtime }) => runtime.deleteMobileSpeechModel(params.modelId)
+  }),
   defineMethod({
     name: 'speech.dictation.setup',
     params: DictationSetup,
diff --git a/src/main/runtime/runtime-rpc.ts b/src/main/runtime/runtime-rpc.ts
index 8628989b3..182f0b493 100644
--- a/src/main/runtime/runtime-rpc.ts
+++ b/src/main/runtime/runtime-rpc.ts
@@ -305,6 +305,7 @@ const MOBILE_RPC_METHOD_ALLOWLIST = new Set([
   'speech.dictation.finish',
   'speech.dictation.setup',
   'speech.dictation.start',
+  'speech.models.delete',
   'speech.models.download',
   'speech.models.list',
   'stats.summary',
diff --git a/src/main/speech/model-manager.test.ts b/src/main/speech/model-manager.test.ts
index dc5815845..aeacac09c 100644
--- a/src/main/speech/model-manager.test.ts
+++ b/src/main/speech/model-manager.test.ts
@@ -1,7 +1,7 @@
 import { createHash } from 'crypto'
-import { mkdtempSync, rmSync, writeFileSync } from 'fs'
+import { existsSync, mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'fs'
 import { tmpdir } from 'os'
-import { join } from 'path'
+import { dirname, join } from 'path'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 import { SPEECH_MODEL_CATALOG } from './model-catalog'
 import { ModelManager } from './model-manager'
@@ -122,6 +122,35 @@ describe('ModelManager', () => {
     }
   })
 
+  it('deletes a ready local model and reports it as not downloaded', async () => {
+    const dir = mkdtempSync(join(tmpdir(), 'orca-model-manager-'))
+    try {
+      const manifest = SPEECH_MODEL_CATALOG.find((model) => model.provider === 'local')
+      expect(manifest?.files).toBeDefined()
+      const manager = new ModelManager(dir)
+      const modelDir = manager.getModelDir(manifest!.id)
+      for (const file of manifest!.files ?? []) {
+        const path = join(modelDir, file)
+        mkdirSync(dirname(path), { recursive: true })
+        writeFileSync(path, 'model file')
+      }
+
+      await expect(manager.getModelState(manifest!.id)).resolves.toEqual({
+        id: manifest!.id,
+        status: 'ready'
+      })
+      await manager.deleteModel(manifest!.id)
+
+      expect(existsSync(modelDir)).toBe(false)
+      await expect(manager.getModelState(manifest!.id)).resolves.toEqual({
+        id: manifest!.id,
+        status: 'not-downloaded'
+      })
+    } finally {
+      rmSync(dir, { recursive: true, force: true })
+    }
+  })
+
   it('aborts an in-flight model download request when cancelled', async () => {
     const dir = mkdtempSync(join(tmpdir(), 'orca-model-manager-'))
     try {
diff --git a/src/main/speech/speech-model-deletion.test.ts b/src/main/speech/speech-model-deletion.test.ts
new file mode 100644
index 000000000..5ce3788d7
--- /dev/null
+++ b/src/main/speech/speech-model-deletion.test.ts
@@ -0,0 +1,134 @@
+import { describe, expect, it, vi } from 'vitest'
+import { getDefaultVoiceSettings } from '../../shared/constants'
+import type { VoiceSettings } from '../../shared/speech-types'
+import type { GlobalSettings } from '../../shared/types'
+import { SPEECH_MODEL_CATALOG } from './model-catalog'
+import { deleteLocalSpeechModel } from './speech-model-deletion'
+
+const localModel = SPEECH_MODEL_CATALOG.find((model) => model.provider === 'local')
+const cloudModel = SPEECH_MODEL_CATALOG.find((model) => model.provider === 'openai')
+
+function makeStore(initialVoice: VoiceSettings) {
+  let voice = initialVoice
+  const updateSettings = vi.fn((updates: Partial<GlobalSettings>) => {
+    if (updates.voice) {
+      voice = updates.voice
+    }
+    return { voice } as GlobalSettings
+  })
+  return {
+    getSettings: vi.fn(() => ({ voice }) as GlobalSettings),
+    updateSettings
+  }
+}
+
+describe('deleteLocalSpeechModel', () => {
+  it('clears the selected local model after deletion succeeds', async () => {
+    expect(localModel).toBeDefined()
+    const calls: string[] = []
+    const voice = { ...getDefaultVoiceSettings(), enabled: true, sttModel: localModel!.id }
+    const store = makeStore(voice)
+    const modelManager = {
+      deleteModel: vi.fn(async () => {
+        calls.push('delete')
+      })
+    }
+    const sttService = {
+      prepareModelForDeletion: vi.fn(async () => {
+        calls.push('prepare')
+      })
+    }
+
+    await deleteLocalSpeechModel({
+      store,
+      modelManager,
+      sttService,
+      modelId: localModel!.id
+    })
+
+    expect(calls).toEqual(['prepare', 'delete'])
+    expect(store.updateSettings).toHaveBeenCalledWith(
+      {
+        voice: {
+          ...voice,
+          sttModel: ''
+        }
+      },
+      { notifyListeners: true }
+    )
+  })
+
+  it('does not clear selection when another client selected a newer model', async () => {
+    expect(localModel).toBeDefined()
+    const voice = { ...getDefaultVoiceSettings(), enabled: true, sttModel: localModel!.id }
+    const store = makeStore(voice)
+    const modelManager = {
+      deleteModel: vi.fn(async () => {
+        store.updateSettings({ voice: { ...voice, sttModel: 'newer-model' } })
+      })
+    }
+    const sttService = { prepareModelForDeletion: vi.fn(async () => {}) }
+
+    await deleteLocalSpeechModel({
+      store,
+      modelManager,
+      sttService,
+      modelId: localModel!.id
+    })
+
+    expect(store.updateSettings).toHaveBeenCalledTimes(1)
+    expect(store.updateSettings).toHaveBeenCalledWith({
+      voice: { ...voice, sttModel: 'newer-model' }
+    })
+  })
+
+  it('leaves settings untouched when deletion fails', async () => {
+    expect(localModel).toBeDefined()
+    const store = makeStore({ ...getDefaultVoiceSettings(), sttModel: localModel!.id })
+    const modelManager = {
+      deleteModel: vi.fn(async () => {
+        throw new Error('permission denied')
+      })
+    }
+    const sttService = { prepareModelForDeletion: vi.fn(async () => {}) }
+
+    await expect(
+      deleteLocalSpeechModel({
+        store,
+        modelManager,
+        sttService,
+        modelId: localModel!.id
+      })
+    ).rejects.toThrow('permission denied')
+
+    expect(store.updateSettings).not.toHaveBeenCalled()
+  })
+
+  it('rejects unknown and cloud models before preparing storage deletion', async () => {
+    expect(cloudModel).toBeDefined()
+    const store = makeStore(getDefaultVoiceSettings())
+    const modelManager = { deleteModel: vi.fn(async () => {}) }
+    const sttService = { prepareModelForDeletion: vi.fn(async () => {}) }
+
+    await expect(
+      deleteLocalSpeechModel({
+        store,
+        modelManager,
+        sttService,
+        modelId: 'missing-model'
+      })
+    ).rejects.toThrow('voice_model_unknown')
+    await expect(
+      deleteLocalSpeechModel({
+        store,
+        modelManager,
+        sttService,
+        modelId: cloudModel!.id
+      })
+    ).rejects.toThrow('voice_model_not_deletable')
+
+    expect(sttService.prepareModelForDeletion).not.toHaveBeenCalled()
+    expect(modelManager.deleteModel).not.toHaveBeenCalled()
+    expect(store.updateSettings).not.toHaveBeenCalled()
+  })
+})
diff --git a/src/main/speech/speech-model-deletion.ts b/src/main/speech/speech-model-deletion.ts
new file mode 100644
index 000000000..32a0c3363
--- /dev/null
+++ b/src/main/speech/speech-model-deletion.ts
@@ -0,0 +1,79 @@
+import { getDefaultVoiceSettings } from '../../shared/constants'
+import type { VoiceSettings } from '../../shared/speech-types'
+import { getCatalogModel, isLocalSpeechModel } from './model-catalog'
+import type { ModelManager } from './model-manager'
+import type { SttService } from './stt-service'
+
+export type SpeechModelDeletionErrorCode =
+  | 'voice_model_unknown'
+  | 'voice_model_not_deletable'
+  | 'voice_model_in_use'
+
+export class SpeechModelDeletionError extends Error {
+  constructor(readonly code: SpeechModelDeletionErrorCode) {
+    super(code)
+    this.name = 'SpeechModelDeletionError'
+  }
+}
+
+type SpeechModelDeletionStore = {
+  getSettings: () => {
+    voice?: VoiceSettings
+  }
+  updateSettings: (
+    updates: {
+      voice: VoiceSettings
+    },
+    options?: { notifyListeners?: boolean; originWebContentsId?: number }
+  ) => unknown
+}
+
+type DeleteLocalSpeechModelArgs = {
+  store: SpeechModelDeletionStore
+  modelManager: Pick<ModelManager, 'deleteModel'>
+  sttService: Pick<SttService, 'prepareModelForDeletion'>
+  modelId: string
+}
+
+export function getSpeechModelDeletionErrorCode(
+  error: unknown
+): SpeechModelDeletionErrorCode | null {
+  if (error instanceof SpeechModelDeletionError) {
+    return error.code
+  }
+  if (error instanceof Error && error.message === 'voice_model_in_use') {
+    return 'voice_model_in_use'
+  }
+  return null
+}
+
+export async function deleteLocalSpeechModel({
+  store,
+  modelManager,
+  sttService,
+  modelId
+}: DeleteLocalSpeechModelArgs): Promise<void> {
+  const manifest = getCatalogModel(modelId)
+  if (!manifest) {
+    throw new SpeechModelDeletionError('voice_model_unknown')
+  }
+  if (!isLocalSpeechModel(manifest)) {
+    throw new SpeechModelDeletionError('voice_model_not_deletable')
+  }
+
+  await sttService.prepareModelForDeletion(modelId)
+  await modelManager.deleteModel(modelId)
+
+  const currentVoice = store.getSettings().voice ?? getDefaultVoiceSettings()
+  if (currentVoice.sttModel === modelId) {
+    store.updateSettings(
+      {
+        voice: {
+          ...currentVoice,
+          sttModel: ''
+        }
+      },
+      { notifyListeners: true }
+    )
+  }
+}
diff --git a/src/main/speech/stt-service.test.ts b/src/main/speech/stt-service.test.ts
index ce0bbc77d..5138831d8 100644
--- a/src/main/speech/stt-service.test.ts
+++ b/src/main/speech/stt-service.test.ts
@@ -239,6 +239,70 @@ describe('SttService', () => {
     expect(worker!.messages.filter((message) => message.type === 'feed')).toHaveLength(0)
   })
 
+  it('rejects deletion prep while the target model is starting', async () => {
+    let resolveModelState: (state: { id: string; status: string }) => void = () => {}
+    const modelStatePromise = new Promise<{ id: string; status: string }>((resolve) => {
+      resolveModelState = resolve
+    })
+    const service = new SttService({
+      getModelState: vi.fn(() => modelStatePromise),
+      getModelDir: vi.fn().mockReturnValue('/tmp/model-a')
+    } as never)
+
+    const startPromise = service.startDictation('model-a', vi.fn(), undefined, 'desktop')
+    await Promise.resolve()
+
+    await expect(service.prepareModelForDeletion('model-a')).rejects.toThrow('voice_model_in_use')
+
+    resolveModelState({ id: 'model-a', status: 'ready' })
+    await startPromise
+    await service.stopDictation('desktop')
+  })
+
+  it('tears down an idle warm worker before deleting the target model', async () => {
+    const service = new SttService({
+      getModelState: vi.fn().mockResolvedValue({ id: 'model-a', status: 'ready' }),
+      getModelDir: vi.fn().mockReturnValue('/tmp/model-a')
+    } as never)
+
+    await service.startDictation('model-a', vi.fn(), undefined, 'desktop')
+    const worker = getLastWorker()
+    expect(worker).toBeDefined()
+    await service.stopDictation('desktop')
+
+    await service.prepareModelForDeletion('model-a')
+
+    expect(worker!.messages.some((message) => message.type === 'teardown')).toBe(true)
+    expect(worker!.terminated).toBe(true)
+    expect(service.isActive()).toBe(false)
+  })
+
+  it('rejects deletion prep when a target warm worker cannot be torn down during another start', async () => {
+    let resolveModelState: (state: { id: string; status: string }) => void = () => {}
+    const secondModelState = new Promise<{ id: string; status: string }>((resolve) => {
+      resolveModelState = resolve
+    })
+    const getModelState = vi
+      .fn()
+      .mockResolvedValueOnce({ id: 'model-a', status: 'ready' })
+      .mockReturnValue(secondModelState)
+    const service = new SttService({
+      getModelState,
+      getModelDir: vi.fn().mockReturnValue('/tmp/model-a')
+    } as never)
+
+    await service.startDictation('model-a', vi.fn(), undefined, 'desktop')
+    await service.stopDictation('desktop')
+    const startOtherModel = service.startDictation('model-b', vi.fn(), undefined, 'desktop')
+    await Promise.resolve()
+
+    await expect(service.prepareModelForDeletion('model-a')).rejects.toThrow('voice_model_in_use')
+
+    resolveModelState({ id: 'model-b', status: 'ready' })
+    await startOtherModel
+    await service.stopDictation('desktop')
+  })
+
   it('uses the OpenAI transcription session without creating a worker', async () => {
     const sink = vi.fn()
     const service = new SttService({
diff --git a/src/main/speech/stt-service.ts b/src/main/speech/stt-service.ts
index f31a3b90e..32e3a16ec 100644
--- a/src/main/speech/stt-service.ts
+++ b/src/main/speech/stt-service.ts
@@ -30,6 +30,7 @@ export class SttService {
   private activeHotwordsFilePath: string | undefined
   private activeOwner: string | null = null
   private startingOwner: string | null = null
+  private startingModelId: string | null = null
   private starting = false
   private canceledOwners = new Set<string>()
   private eventSink: SttEventSink | null = null
@@ -59,6 +60,7 @@ export class SttService {
     }
     this.starting = true
     this.startingOwner = owner
+    this.startingModelId = modelId
     this.clearIdleTeardownTimer()
 
     try {
@@ -71,6 +73,7 @@ export class SttService {
     } finally {
       this.starting = false
       this.startingOwner = null
+      this.startingModelId = null
       this.canceledOwners.delete(owner)
     }
   }
@@ -382,6 +385,18 @@ export class SttService {
     return this.activeModelId
   }
 
+  async prepareModelForDeletion(modelId: string): Promise<void> {
+    if (this.startingModelId === modelId || (this.activeOwner && this.activeModelId === modelId)) {
+      throw new Error('voice_model_in_use')
+    }
+    if (this.worker && this.activeModelId === modelId) {
+      await this.teardownIdleWorker({ ignoreTerminateErrors: false })
+      if (this.worker && this.activeModelId === modelId) {
+        throw new Error('voice_model_in_use')
+      }
+    }
+  }
+
   private getWorkerPath(): string {
     if (app.isPackaged) {
       return join(process.resourcesPath, 'app.asar', 'out', 'main', 'stt-worker.js')
@@ -407,16 +422,24 @@ export class SttService {
     this.idleTeardownTimer.unref?.()
   }
 
-  private async teardownIdleWorker(): Promise<void> {
+  private async teardownIdleWorker(
+    options: { ignoreTerminateErrors?: boolean } = { ignoreTerminateErrors: true }
+  ): Promise<void> {
     this.clearIdleTeardownTimer()
     if (!this.worker || this.activeOwner || this.startingOwner) {
       return
     }
     const worker = this.worker
     worker.postMessage({ type: 'teardown' })
+    try {
+      await worker.terminate()
+    } catch (error) {
+      if (!options.ignoreTerminateErrors) {
+        throw error
+      }
+    }
     this.cleanupActiveWorkerLifecycleListeners()
     worker.removeAllListeners()
-    await worker.terminate().catch(() => undefined)
     if (this.worker === worker) {
       this.worker = null
       this.activeModelId = null
diff --git a/src/renderer/src/components/settings/VoiceSpeechModelSection.test.tsx b/src/renderer/src/components/settings/VoiceSpeechModelSection.test.tsx
new file mode 100644
index 000000000..b6e061994
--- /dev/null
+++ b/src/renderer/src/components/settings/VoiceSpeechModelSection.test.tsx
@@ -0,0 +1,221 @@
+// @vitest-environment happy-dom
+
+import { act } from 'react'
+import type { ReactNode } from 'react'
+import { createRoot, type Root } from 'react-dom/client'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import type { SpeechModelManifest, SpeechModelState } from '../../../../shared/speech-types'
+import { getDefaultVoiceSettings } from '../../../../shared/constants'
+
+const toastErrorMock = vi.hoisted(() => vi.fn())
+
+vi.mock('sonner', () => ({
+  toast: {
+    error: toastErrorMock
+  }
+}))
+
+vi.mock('@/i18n/i18n', () => ({
+  translate: (_key: string, fallback: string, values?: Record<string, string>) =>
+    values ? fallback.replace('{{value0}}', values.value0) : fallback
+}))
+
+vi.mock('../ui/dropdown-menu', () => ({
+  DropdownMenu: ({ children }: { children: ReactNode }) => <div>{children}</div>,
+  DropdownMenuTrigger: ({ children }: { children: ReactNode }) => <>{children}</>,
+  DropdownMenuContent: ({ children }: { children: ReactNode }) => <div>{children}</div>,
+  DropdownMenuItem: ({
+    children,
+    disabled,
+    onSelect,
+    className
+  }: {
+    children: ReactNode
+    disabled?: boolean
+    onSelect?: () => void
+    className?: string
+  }) => (
+    <div
+      className={className}
+      aria-disabled={disabled}
+      role="option"
+      onClick={() => {
+        if (!disabled) {
+          onSelect?.()
+        }
+      }}
+    >
+      {children}
+    </div>
+  )
+}))
+
+import { VoiceSpeechModelSection } from './VoiceSpeechModelSection'
+
+const localModel: SpeechModelManifest = {
+  id: 'model-a',
+  label: 'Local Model',
+  description: 'Runs offline',
+  provider: 'local',
+  language: 'en',
+  type: 'transducer',
+  streaming: true,
+  sampleRate: 16000,
+  sizeBytes: 123_000_000,
+  files: ['encoder.onnx']
+}
+
+const secondLocalModel: SpeechModelManifest = {
+  ...localModel,
+  id: 'model-b',
+  label: 'Second Local Model'
+}
+
+function renderSection(args: {
+  deleteModel: (modelId: string) => Promise<void>
+  catalog?: SpeechModelManifest[]
+  modelStates?: SpeechModelState[]
+  refreshModelStates?: () => void
+}): { container: HTMLDivElement; root: Root } {
+  Object.assign(window, {
+    api: {
+      speech: {
+        deleteModel: vi.fn(args.deleteModel),
+        downloadModel: vi.fn()
+      }
+    }
+  })
+
+  const container = document.createElement('div')
+  document.body.appendChild(container)
+  const root = createRoot(container)
+  const voiceSettings = { ...getDefaultVoiceSettings(), enabled: true, sttModel: localModel.id }
+  const catalog = args.catalog ?? [localModel]
+  const modelStates = args.modelStates ?? [{ id: localModel.id, status: 'ready' }]
+  act(() => {
+    root.render(
+      <VoiceSpeechModelSection
+        voiceSettings={voiceSettings}
+        catalog={catalog}
+        modelStates={modelStates}
+        onUpdateVoiceSettings={vi.fn()}
+        onOpenOpenAiDialog={vi.fn()}
+        onRefreshModelStates={args.refreshModelStates ?? vi.fn()}
+      />
+    )
+  })
+
+  return { container, root }
+}
+
+describe('VoiceSpeechModelSection', () => {
+  beforeEach(() => {
+    toastErrorMock.mockReset()
+  })
+
+  afterEach(() => {
+    document.body.innerHTML = ''
+    vi.unstubAllGlobals()
+  })
+
+  it('shows delete for the selected ready local row and refreshes after success', async () => {
+    let resolveDelete: () => void = () => {}
+    const refreshModelStates = vi.fn()
+    const { container, root } = renderSection({
+      refreshModelStates,
+      deleteModel: () =>
+        new Promise<void>((resolve) => {
+          resolveDelete = resolve
+        })
+    })
+    const deleteButton = container.querySelector<HTMLButtonElement>(
+      'button[aria-label="Delete Local Model"]'
+    )
+
+    expect(deleteButton).not.toBeNull()
+    await act(async () => {
+      deleteButton!.dispatchEvent(new MouseEvent('click', { bubbles: true }))
+    })
+    expect(window.api.speech.deleteModel).toHaveBeenCalledWith(localModel.id)
+    expect(deleteButton!.disabled).toBe(true)
+
+    await act(async () => {
+      deleteButton!.dispatchEvent(new MouseEvent('click', { bubbles: true }))
+      resolveDelete()
+      await Promise.resolve()
+    })
+
+    expect(window.api.speech.deleteModel).toHaveBeenCalledTimes(1)
+    expect(refreshModelStates).toHaveBeenCalledTimes(1)
+    root.unmount()
+  })
+
+  it('keeps another row delete disabled until its own request finishes', async () => {
+    const deleteResolvers = new Map<string, () => void>()
+    const refreshModelStates = vi.fn()
+    const { container, root } = renderSection({
+      refreshModelStates,
+      catalog: [localModel, secondLocalModel],
+      modelStates: [
+        { id: localModel.id, status: 'ready' },
+        { id: secondLocalModel.id, status: 'ready' }
+      ],
+      deleteModel: (modelId) =>
+        new Promise<void>((resolve) => {
+          deleteResolvers.set(modelId, resolve)
+        })
+    })
+    const firstDeleteButton = container.querySelector<HTMLButtonElement>(
+      'button[aria-label="Delete Local Model"]'
+    )
+    const secondDeleteButton = container.querySelector<HTMLButtonElement>(
+      'button[aria-label="Delete Second Local Model"]'
+    )
+
+    await act(async () => {
+      firstDeleteButton!.dispatchEvent(new MouseEvent('click', { bubbles: true }))
+    })
+    await act(async () => {
+      secondDeleteButton!.dispatchEvent(new MouseEvent('click', { bubbles: true }))
+    })
+
+    expect(firstDeleteButton!.disabled).toBe(true)
+    expect(secondDeleteButton!.disabled).toBe(true)
+
+    await act(async () => {
+      deleteResolvers.get(localModel.id)!()
+      await Promise.resolve()
+    })
+
+    expect(firstDeleteButton!.disabled).toBe(false)
+    expect(secondDeleteButton!.disabled).toBe(true)
+
+    await act(async () => {
+      deleteResolvers.get(secondLocalModel.id)!()
+      await Promise.resolve()
+    })
+
+    expect(refreshModelStates).toHaveBeenCalledTimes(2)
+    root.unmount()
+  })
+
+  it('shows the existing error toast when selected-row deletion fails', async () => {
+    const refreshModelStates = vi.fn()
+    const { container, root } = renderSection({
+      refreshModelStates,
+      deleteModel: () => Promise.reject(new Error('in use'))
+    })
+    const deleteButton = container.querySelector<HTMLButtonElement>(
+      'button[aria-label="Delete Local Model"]'
+    )
+
+    await act(async () => {
+      deleteButton!.dispatchEvent(new MouseEvent('click', { bubbles: true }))
+      await Promise.resolve()
+    })
+
+    expect(toastErrorMock).toHaveBeenCalledWith('Failed to delete model.')
+    expect(refreshModelStates).not.toHaveBeenCalled()
+    root.unmount()
+  })
+})
diff --git a/src/renderer/src/components/settings/VoiceSpeechModelSection.tsx b/src/renderer/src/components/settings/VoiceSpeechModelSection.tsx
index 870b498b5..0380bbc0c 100644
--- a/src/renderer/src/components/settings/VoiceSpeechModelSection.tsx
+++ b/src/renderer/src/components/settings/VoiceSpeechModelSection.tsx
@@ -1,3 +1,4 @@
+import { useState } from 'react'
 import { toast } from 'sonner'
 import type { VoiceSettings } from '../../../../shared/speech-types'
 import type { SpeechModelManifest, SpeechModelState } from '../../../../shared/speech-types'
@@ -29,6 +30,7 @@ export function VoiceSpeechModelSection({
   onOpenOpenAiDialog,
   onRefreshModelStates
 }: VoiceSpeechModelSectionProps): React.JSX.Element {
+  const [pendingDeleteModelIds, setPendingDeleteModelIds] = useState<Set<string>>(() => new Set())
   const getModelState = (id: string): SpeechModelState | undefined =>
     modelStates.find((s) => s.id === id)
 
@@ -73,6 +75,7 @@ export function VoiceSpeechModelSection({
               mState?.status === 'downloading' || mState?.status === 'extracting'
             const isActive = voiceSettings.sttModel === manifest.id
             const isCloud = manifest.provider === 'openai'
+            const deletePending = pendingDeleteModelIds.has(manifest.id)
             const sizeMb = manifest.sizeBytes ? Math.round(manifest.sizeBytes / 1_000_000) : null
 
             return (
@@ -146,10 +149,34 @@ export function VoiceSpeechModelSection({
                     {manifest.description}
                   </p>
                 </div>
-                {!isCloud && isReady && !isActive ? (
-                  <button
-                    onClick={(e) => {
-                      e.stopPropagation()
+                {!isCloud && isReady ? (
+                  <Button
+                    type="button"
+                    variant="ghost"
+                    size="icon-xs"
+                    aria-label={translate(
+                      'auto.components.settings.VoicePane.6fa734ed95',
+                      'Delete {{value0}}',
+                      {
+                        value0: manifest.label
+                      }
+                    )}
+                    disabled={deletePending}
+                    onMouseDown={(event) => {
+                      event.preventDefault()
+                      event.stopPropagation()
+                    }}
+                    onClick={(event) => {
+                      event.preventDefault()
+                      event.stopPropagation()
+                      if (deletePending) {
+                        return
+                      }
+                      setPendingDeleteModelIds((prev) => {
+                        const next = new Set(prev)
+                        next.add(manifest.id)
+                        return next
+                      })
                       void window.api.speech
                         .deleteModel(manifest.id)
                         .then(onRefreshModelStates)
@@ -161,11 +188,22 @@ export function VoiceSpeechModelSection({
                             )
                           )
                         )
+                        .finally(() =>
+                          setPendingDeleteModelIds((prev) => {
+                            const next = new Set(prev)
+                            next.delete(manifest.id)
+                            return next
+                          })
+                        )
                     }}
-                    className="shrink-0 p-1 text-muted-foreground can-hover:opacity-0 group-hover:opacity-100 hover:text-destructive transition-all rounded"
+                    className="shrink-0 text-muted-foreground can-hover:opacity-0 group-hover:opacity-100 hover:text-destructive disabled:opacity-60 disabled:hover:text-muted-foreground"
                   >
-                    <Trash2 className="size-3" />
-                  </button>
+                    {deletePending ? (
+                      <Loader2 className="size-3 animate-spin" />
+                    ) : (
+                      <Trash2 className="size-3" />
+                    )}
+                  </Button>
                 ) : !isCloud && !isReady && !isDownloading ? (
                   <span className="shrink-0 p-1 text-muted-foreground can-hover:opacity-0 group-hover:opacity-100 transition-opacity">
                     <Download className="size-3" />
diff --git a/src/renderer/src/i18n/locales/en.json b/src/renderer/src/i18n/locales/en.json
index 25c6577d3..be7d84706 100644
--- a/src/renderer/src/i18n/locales/en.json
+++ b/src/renderer/src/i18n/locales/en.json
@@ -6277,6 +6277,7 @@
           "c2c64b24d0": "Reset"
         },
         "VoicePane": {
+          "6fa734ed95": "Delete {{value0}}",
           "68de13f72c": "Failed to delete model.",
           "1ba81c0ff0": "recommended",
           "cfde55c7b0": "Failed to download model.",
diff --git a/src/renderer/src/i18n/locales/es.json b/src/renderer/src/i18n/locales/es.json
index 0866938eb..a0a53a4a9 100644
--- a/src/renderer/src/i18n/locales/es.json
+++ b/src/renderer/src/i18n/locales/es.json
@@ -6268,7 +6268,8 @@
           "ad5d036ecc": "No se pudo solicitar permiso del micrófono. El dictado de voz no estaba habilitado.",
           "f9a9cf6928": "Se requiere permiso del micrófono antes de habilitar el dictado de voz.",
           "1eac933202": "Se abrió Privacidad y seguridad de macOS. Habilite el dictado nuevamente después de otorgar acceso.",
-          "cd9fe37556": "Permiso de micrófono concedido"
+          "cd9fe37556": "Permiso de micrófono concedido",
+          "6fa734ed95": "Delete {{value0}}"
         },
         "WorktreeSymlinksSection": {
           "1c1e35b219": "Quitar {{value0}}",
diff --git a/src/renderer/src/i18n/locales/ja.json b/src/renderer/src/i18n/locales/ja.json
index 50f47f047..e7ce8ab1d 100644
--- a/src/renderer/src/i18n/locales/ja.json
+++ b/src/renderer/src/i18n/locales/ja.json
@@ -6290,7 +6290,8 @@
           "ad5d036ecc": "マイクの許可を要求できませんでした。音声ディクテーションが有効になっていませんでした。",
           "f9a9cf6928": "音声ディクテーションを有効にする前に、マイクの許可が必要です。",
           "1eac933202": "macOS のプライバシーとセキュリティを開きました。アクセスを許可した後、ディクテーションを再度有効にします。",
-          "cd9fe37556": "マイクの許可が与えられました"
+          "cd9fe37556": "マイクの許可が与えられました",
+          "6fa734ed95": "Delete {{value0}}"
         },
         "WorktreeSymlinksSection": {
           "1c1e35b219": "{{value0}} を削除",
diff --git a/src/renderer/src/i18n/locales/ko.json b/src/renderer/src/i18n/locales/ko.json
index 919609547..aafb68a9a 100644
--- a/src/renderer/src/i18n/locales/ko.json
+++ b/src/renderer/src/i18n/locales/ko.json
@@ -6253,7 +6253,8 @@
           "ad5d036ecc": "마이크 권한을 요청할 수 없습니다. 음성 받아쓰기가 활성화되지 않았습니다.",
           "f9a9cf6928": "음성 받아쓰기를 활성화하려면 마이크 권한이 필요합니다.",
           "1eac933202": "macOS 개인 정보 보호 및 보안을 열었습니다. 액세스 권한을 부여한 후 받아쓰기를 다시 활성화하세요.",
-          "cd9fe37556": "마이크 권한이 부여되었습니다."
+          "cd9fe37556": "마이크 권한이 부여되었습니다.",
+          "6fa734ed95": "Delete {{value0}}"
         },
         "WorktreeSymlinksSection": {
           "1c1e35b219": "{{value0}} 제거",
diff --git a/src/renderer/src/i18n/locales/zh.json b/src/renderer/src/i18n/locales/zh.json
index 1b1569758..1cea91547 100644
--- a/src/renderer/src/i18n/locales/zh.json
+++ b/src/renderer/src/i18n/locales/zh.json
@@ -6253,7 +6253,8 @@
           "ad5d036ecc": "无法请求麦克风权限。未启用语音听写。",
           "f9a9cf6928": "启用语音听写之前需要麦克风许可。",
           "1eac933202": "打开 macOS 隐私和安全。授予访问权限后再次启用听写。",
-          "cd9fe37556": "已授予麦克风权限"
+          "cd9fe37556": "已授予麦克风权限",
+          "6fa734ed95": "Delete {{value0}}"
         },
         "WorktreeSymlinksSection": {
           "1c1e35b219": "删除 {{value0}}",