diff --git a/src/extension-host/media-runtime-backends.test.ts b/src/extension-host/media-runtime-backends.test.ts
new file mode 100644
index 00000000000..d5acac9661b
--- /dev/null
+++ b/src/extension-host/media-runtime-backends.test.ts
@@ -0,0 +1,60 @@
+import { describe, expect, it } from "vitest";
+import {
+  buildExtensionHostMediaRuntimeSelectorKeys,
+  listExtensionHostMediaAutoRuntimeBackendSeedIds,
+  listExtensionHostMediaRuntimeBackendIds,
+  listExtensionHostMediaUnderstandingProviders,
+  normalizeExtensionHostMediaProviderId,
+  resolveExtensionHostMediaRuntimeDefaultModelMetadata,
+} from "./media-runtime-backends.js";
+
+describe("extension host media runtime backends", () => {
+  it("publishes the built-in media providers once", () => {
+    const providers = listExtensionHostMediaUnderstandingProviders();
+
+    expect(providers.some((provider) => provider.id === "openai")).toBe(true);
+    expect(providers.some((provider) => provider.id === "deepgram")).toBe(true);
+  });
+
+  it("keeps media-specific provider normalization and selector aliases", () => {
+    expect(normalizeExtensionHostMediaProviderId("gemini")).toBe("google");
+    expect(buildExtensionHostMediaRuntimeSelectorKeys("google")).toEqual(["google", "gemini"]);
+  });
+
+  it("keeps auto-seeded runtime backends ordered ahead of the rest", () => {
+    expect(listExtensionHostMediaAutoRuntimeBackendSeedIds("image")).toEqual([
+      "openai",
+      "anthropic",
+      "google",
+      "minimax",
+      "minimax-portal",
+      "zai",
+    ]);
+    expect(listExtensionHostMediaRuntimeBackendIds("audio").slice(0, 3)).toEqual([
+      "openai",
+      "groq",
+      "deepgram",
+    ]);
+    expect(listExtensionHostMediaRuntimeBackendIds("image").slice(0, 4)).toEqual([
+      "openai",
+      "anthropic",
+      "google",
+      "minimax",
+    ]);
+  });
+
+  it("keeps default-model metadata with the shared backend definitions", () => {
+    expect(
+      resolveExtensionHostMediaRuntimeDefaultModelMetadata({
+        capability: "image",
+        backendId: "openai",
+      }),
+    ).toBe("gpt-5-mini");
+    expect(
+      resolveExtensionHostMediaRuntimeDefaultModelMetadata({
+        capability: "video",
+        backendId: "openai",
+      }),
+    ).toBeUndefined();
+  });
+});
diff --git a/src/extension-host/media-runtime-backends.ts b/src/extension-host/media-runtime-backends.ts
new file mode 100644
index 00000000000..22eccc8bb04
--- /dev/null
+++ b/src/extension-host/media-runtime-backends.ts
@@ -0,0 +1,118 @@
+import { normalizeProviderId } from "../agents/provider-id.js";
+import {
+  AUTO_AUDIO_KEY_PROVIDERS,
+  AUTO_IMAGE_KEY_PROVIDERS,
+  AUTO_VIDEO_KEY_PROVIDERS,
+  DEFAULT_AUDIO_MODELS,
+  DEFAULT_IMAGE_MODELS,
+} from "../media-understanding/defaults.js";
+import { anthropicProvider } from "../media-understanding/providers/anthropic/index.js";
+import { deepgramProvider } from "../media-understanding/providers/deepgram/index.js";
+import { googleProvider } from "../media-understanding/providers/google/index.js";
+import { groqProvider } from "../media-understanding/providers/groq/index.js";
+import {
+  minimaxPortalProvider,
+  minimaxProvider,
+} from "../media-understanding/providers/minimax/index.js";
+import { mistralProvider } from "../media-understanding/providers/mistral/index.js";
+import { moonshotProvider } from "../media-understanding/providers/moonshot/index.js";
+import { openaiProvider } from "../media-understanding/providers/openai/index.js";
+import { zaiProvider } from "../media-understanding/providers/zai/index.js";
+import type {
+  MediaUnderstandingCapability,
+  MediaUnderstandingProvider,
+} from "../media-understanding/types.js";
+
+const EXTENSION_HOST_MEDIA_UNDERSTANDING_PROVIDERS: readonly MediaUnderstandingProvider[] = [
+  groqProvider,
+  openaiProvider,
+  googleProvider,
+  anthropicProvider,
+  minimaxProvider,
+  minimaxPortalProvider,
+  moonshotProvider,
+  mistralProvider,
+  zaiProvider,
+  deepgramProvider,
+];
+
+const EXTENSION_HOST_MEDIA_AUTO_RUNTIME_BACKEND_IDS: Record<
+  MediaUnderstandingCapability,
+  readonly string[]
+> = {
+  audio: AUTO_AUDIO_KEY_PROVIDERS,
+  image: AUTO_IMAGE_KEY_PROVIDERS,
+  video: AUTO_VIDEO_KEY_PROVIDERS,
+};
+
+export function listExtensionHostMediaUnderstandingProviders(): readonly MediaUnderstandingProvider[] {
+  return EXTENSION_HOST_MEDIA_UNDERSTANDING_PROVIDERS;
+}
+
+export function normalizeExtensionHostMediaProviderId(id: string): string {
+  const normalized = normalizeProviderId(id);
+  if (normalized === "gemini") {
+    return "google";
+  }
+  return normalized;
+}
+
+export function buildExtensionHostMediaRuntimeSelectorKeys(providerId: string): readonly string[] {
+  const normalized = normalizeExtensionHostMediaProviderId(providerId);
+  if (normalized === "google") {
+    return [providerId, "gemini"];
+  }
+  return normalized === providerId ? [providerId] : [providerId, normalized];
+}
+
+export function listExtensionHostMediaAutoRuntimeBackendSeedIds(
+  capability: MediaUnderstandingCapability,
+): readonly string[] {
+  return EXTENSION_HOST_MEDIA_AUTO_RUNTIME_BACKEND_IDS[capability];
+}
+
+export function listExtensionHostMediaRuntimeBackendIds(
+  capability: MediaUnderstandingCapability,
+): readonly string[] {
+  const ordered: string[] = [];
+  const seen = new Set<string>();
+  const pushProvider = (provider: MediaUnderstandingProvider | undefined) => {
+    if (!provider || !(provider.capabilities ?? []).includes(capability)) {
+      return;
+    }
+    const normalized = normalizeExtensionHostMediaProviderId(provider.id);
+    if (seen.has(normalized)) {
+      return;
+    }
+    seen.add(normalized);
+    ordered.push(normalized);
+  };
+
+  const providersById = new Map(
+    listExtensionHostMediaUnderstandingProviders().map((provider) => [
+      normalizeExtensionHostMediaProviderId(provider.id),
+      provider,
+    ]),
+  );
+
+  for (const providerId of listExtensionHostMediaAutoRuntimeBackendSeedIds(capability)) {
+    pushProvider(providersById.get(normalizeExtensionHostMediaProviderId(providerId)));
+  }
+  for (const provider of providersById.values()) {
+    pushProvider(provider);
+  }
+  return ordered;
+}
+
+export function resolveExtensionHostMediaRuntimeDefaultModelMetadata(params: {
+  capability: MediaUnderstandingCapability;
+  backendId: string;
+}): string | undefined {
+  if (params.capability === "audio") {
+    return DEFAULT_AUDIO_MODELS[params.backendId];
+  }
+  if (params.capability === "image") {
+    return DEFAULT_IMAGE_MODELS[params.backendId];
+  }
+  return undefined;
+}
diff --git a/src/extension-host/media-runtime-registry.ts b/src/extension-host/media-runtime-registry.ts
index 7b6479dab61..00a15ed4a1a 100644
--- a/src/extension-host/media-runtime-registry.ts
+++ b/src/extension-host/media-runtime-registry.ts
@@ -1,49 +1,21 @@
-import { normalizeProviderId } from "../agents/provider-id.js";
-import { anthropicProvider } from "../media-understanding/providers/anthropic/index.js";
-import { deepgramProvider } from "../media-understanding/providers/deepgram/index.js";
-import { googleProvider } from "../media-understanding/providers/google/index.js";
-import { groqProvider } from "../media-understanding/providers/groq/index.js";
-import {
-  minimaxPortalProvider,
-  minimaxProvider,
-} from "../media-understanding/providers/minimax/index.js";
-import { mistralProvider } from "../media-understanding/providers/mistral/index.js";
-import { moonshotProvider } from "../media-understanding/providers/moonshot/index.js";
-import { openaiProvider } from "../media-understanding/providers/openai/index.js";
-import { zaiProvider } from "../media-understanding/providers/zai/index.js";
 import type { MediaUnderstandingProvider } from "../media-understanding/types.js";
-
-const EXTENSION_HOST_MEDIA_PROVIDERS: readonly MediaUnderstandingProvider[] = [
-  groqProvider,
-  openaiProvider,
-  googleProvider,
-  anthropicProvider,
-  minimaxProvider,
-  minimaxPortalProvider,
-  moonshotProvider,
-  mistralProvider,
-  zaiProvider,
-  deepgramProvider,
-];
+import {
+  listExtensionHostMediaUnderstandingProviders,
+  normalizeExtensionHostMediaProviderId,
+} from "./media-runtime-backends.js";
 
 export type ExtensionHostMediaUnderstandingProviderRegistry = Map<
   string,
   MediaUnderstandingProvider
 >;
 
-export function normalizeExtensionHostMediaProviderId(id: string): string {
-  const normalized = normalizeProviderId(id);
-  if (normalized === "gemini") {
-    return "google";
-  }
-  return normalized;
-}
+export { normalizeExtensionHostMediaProviderId } from "./media-runtime-backends.js";
 
 export function buildExtensionHostMediaUnderstandingRegistry(
   overrides?: Record<string, MediaUnderstandingProvider>,
 ): ExtensionHostMediaUnderstandingProviderRegistry {
   const registry: ExtensionHostMediaUnderstandingProviderRegistry = new Map();
-  for (const provider of EXTENSION_HOST_MEDIA_PROVIDERS) {
+  for (const provider of listExtensionHostMediaUnderstandingProviders()) {
     registry.set(normalizeExtensionHostMediaProviderId(provider.id), provider);
   }
   if (!overrides) {
diff --git a/src/extension-host/runtime-backend-catalog.test.ts b/src/extension-host/runtime-backend-catalog.test.ts
index eddaf5633ac..ce45d62291a 100644
--- a/src/extension-host/runtime-backend-catalog.test.ts
+++ b/src/extension-host/runtime-backend-catalog.test.ts
@@ -15,36 +15,33 @@ vi.mock("./embedding-runtime-backends.js", () => ({
   ),
 }));
 
-vi.mock("./media-runtime-registry.js", () => ({
-  buildExtensionHostMediaUnderstandingRegistry: vi.fn(
-    () =>
-      new Map([
-        [
-          "openai",
-          {
-            id: "openai",
-            capabilities: ["image", "video"],
-          },
-        ],
-        [
-          "google",
-          {
-            id: "google",
-            capabilities: ["image"],
-          },
-        ],
-        [
-          "deepgram",
-          {
-            id: "deepgram",
-            capabilities: ["audio"],
-          },
-        ],
-      ]),
+vi.mock("./media-runtime-backends.js", () => ({
+  buildExtensionHostMediaRuntimeSelectorKeys: vi.fn((id: string) =>
+    id === "google" ? ["google", "gemini"] : [id],
+  ),
+  listExtensionHostMediaAutoRuntimeBackendSeedIds: vi.fn(
+    (capability: "audio" | "image" | "video") =>
+      ({
+        audio: ["deepgram"],
+        image: ["openai", "google"],
+        video: ["openai"],
+      })[capability],
+  ),
+  listExtensionHostMediaRuntimeBackendIds: vi.fn(
+    (capability: "audio" | "image" | "video") =>
+      ({
+        audio: ["deepgram"],
+        image: ["openai", "google"],
+        video: ["openai"],
+      })[capability],
   ),
   normalizeExtensionHostMediaProviderId: vi.fn((id: string) =>
     id.trim().toLowerCase() === "gemini" ? "google" : id.trim().toLowerCase(),
   ),
+  resolveExtensionHostMediaRuntimeDefaultModelMetadata: vi.fn(
+    (params: { capability: "audio" | "image" | "video"; backendId: string }) =>
+      params.capability === "image" && params.backendId === "openai" ? "gpt-5-mini" : undefined,
+  ),
 }));
 
 vi.mock("./tts-runtime-backends.js", () => ({
diff --git a/src/extension-host/runtime-backend-catalog.ts b/src/extension-host/runtime-backend-catalog.ts
index 3ccebd9fcb7..821ebdc4348 100644
--- a/src/extension-host/runtime-backend-catalog.ts
+++ b/src/extension-host/runtime-backend-catalog.ts
@@ -1,11 +1,4 @@
 import type { TtsProvider } from "../config/types.tts.js";
-import {
-  AUTO_AUDIO_KEY_PROVIDERS,
-  AUTO_IMAGE_KEY_PROVIDERS,
-  AUTO_VIDEO_KEY_PROVIDERS,
-  DEFAULT_AUDIO_MODELS,
-  DEFAULT_IMAGE_MODELS,
-} from "../media-understanding/defaults.js";
 import type { MediaUnderstandingCapability } from "../media-understanding/types.js";
 import {
   EXTENSION_HOST_EMBEDDING_RUNTIME_BACKEND_IDS,
@@ -13,9 +6,12 @@ import {
 } from "./embedding-runtime-backends.js";
 import type { EmbeddingProviderId } from "./embedding-runtime-types.js";
 import {
-  buildExtensionHostMediaUnderstandingRegistry,
+  buildExtensionHostMediaRuntimeSelectorKeys,
+  listExtensionHostMediaAutoRuntimeBackendSeedIds,
+  listExtensionHostMediaRuntimeBackendIds as listExtensionHostMediaRuntimeBackendIdsFromDefinitions,
   normalizeExtensionHostMediaProviderId,
-} from "./media-runtime-registry.js";
+  resolveExtensionHostMediaRuntimeDefaultModelMetadata,
+} from "./media-runtime-backends.js";
 import { listExtensionHostTtsRuntimeBackends } from "./tts-runtime-backends.js";
 
 export const EXTENSION_HOST_RUNTIME_BACKEND_FAMILY = "capability.runtime-backend";
@@ -46,15 +42,6 @@ type ExtensionHostMediaRuntimeSubsystemId = Extract<
   "media.audio" | "media.image" | "media.video"
 >;
 
-const EXTENSION_HOST_MEDIA_AUTO_PROVIDER_IDS: Record<
-  MediaUnderstandingCapability,
-  readonly string[]
-> = {
-  audio: AUTO_AUDIO_KEY_PROVIDERS,
-  image: AUTO_IMAGE_KEY_PROVIDERS,
-  video: AUTO_VIDEO_KEY_PROVIDERS,
-};
-
 function buildRuntimeBackendCatalogId(
   subsystemId: ExtensionHostRuntimeBackendSubsystemId,
   backendId: string,
@@ -74,52 +61,6 @@ function mapMediaCapabilityToSubsystem(
   return "media.image";
 }
 
-function buildMediaSelectorKeys(providerId: string): readonly string[] {
-  const normalized = normalizeExtensionHostMediaProviderId(providerId);
-  if (normalized === "google") {
-    return [providerId, "gemini"];
-  }
-  return normalized === providerId ? [providerId] : [providerId, normalized];
-}
-
-function buildExtensionHostMediaRuntimeProviderIds(
-  capability: MediaUnderstandingCapability,
-): readonly string[] {
-  const registry = buildExtensionHostMediaUnderstandingRegistry();
-  const ordered: string[] = [];
-  const seen = new Set<string>();
-  const pushProvider = (providerId: string) => {
-    const normalized = normalizeExtensionHostMediaProviderId(providerId);
-    const provider = registry.get(normalized);
-    if (!provider || seen.has(normalized) || !(provider.capabilities ?? []).includes(capability)) {
-      return;
-    }
-    seen.add(normalized);
-    ordered.push(normalized);
-  };
-
-  for (const providerId of EXTENSION_HOST_MEDIA_AUTO_PROVIDER_IDS[capability]) {
-    pushProvider(providerId);
-  }
-  for (const provider of registry.values()) {
-    pushProvider(provider.id);
-  }
-  return ordered;
-}
-
-function resolveExtensionHostMediaRuntimeDefaultModelFromDefaults(params: {
-  capability: MediaUnderstandingCapability;
-  backendId: string;
-}): string | undefined {
-  if (params.capability === "audio") {
-    return DEFAULT_AUDIO_MODELS[params.backendId];
-  }
-  if (params.capability === "image") {
-    return DEFAULT_IMAGE_MODELS[params.backendId];
-  }
-  return undefined;
-}
-
 export function listExtensionHostEmbeddingRuntimeBackendCatalogEntries(): readonly ExtensionHostRuntimeBackendCatalogEntry[] {
   return EXTENSION_HOST_EMBEDDING_RUNTIME_BACKEND_IDS.map((backendId, defaultRank) => ({
     id: buildRuntimeBackendCatalogId("embedding", backendId),
@@ -144,29 +85,26 @@ export function listExtensionHostEmbeddingRemoteRuntimeBackendIds(): readonly Em
 
 export function listExtensionHostMediaRuntimeBackendCatalogEntries(): readonly ExtensionHostRuntimeBackendCatalogEntry[] {
   const entries: ExtensionHostRuntimeBackendCatalogEntry[] = [];
-  const registry = buildExtensionHostMediaUnderstandingRegistry();
   for (const capability of ["audio", "image", "video"] as const) {
-    const providerIds = buildExtensionHostMediaRuntimeProviderIds(capability);
+    const providerIds = listExtensionHostMediaRuntimeBackendIdsFromDefinitions(capability);
     for (const [defaultRank, providerId] of providerIds.entries()) {
-      const provider = registry.get(providerId);
-      if (!provider) {
-        continue;
-      }
-      const defaultModel = resolveExtensionHostMediaRuntimeDefaultModelFromDefaults({
+      const defaultModel = resolveExtensionHostMediaRuntimeDefaultModelMetadata({
         capability,
         backendId: providerId,
       });
       entries.push({
-        id: buildRuntimeBackendCatalogId(mapMediaCapabilityToSubsystem(capability), provider.id),
+        id: buildRuntimeBackendCatalogId(mapMediaCapabilityToSubsystem(capability), providerId),
         family: EXTENSION_HOST_RUNTIME_BACKEND_FAMILY,
         subsystemId: mapMediaCapabilityToSubsystem(capability),
-        backendId: provider.id,
+        backendId: providerId,
         source: "builtin",
         defaultRank,
-        selectorKeys: buildMediaSelectorKeys(provider.id),
+        selectorKeys: buildExtensionHostMediaRuntimeSelectorKeys(providerId),
         capabilities: [capability],
         metadata: {
-          autoSelectable: EXTENSION_HOST_MEDIA_AUTO_PROVIDER_IDS[capability].includes(provider.id),
+          autoSelectable: listExtensionHostMediaAutoRuntimeBackendSeedIds(capability).includes(
+            normalizeExtensionHostMediaProviderId(providerId),
+          ),
           ...(defaultModel ? { defaultModel } : {}),
         },
       });