fix(core): support cross-auth fast side queries (#4117)

* fix(core): support cross-auth fast side queries

* refactor(core): hoist resolveForModel selector and refresh side-query docs

Compute the model selector once at the top of `resolveForModel` and pass
it through to `createContentGeneratorForModel` and
`resolveModelAcrossAuthTypes`. This eliminates the redundant selector
resolution that happened up to five times per cross-auth side query
(once per call, plus once inside each downstream helper).

Also update the JSDoc for `SideQueryJsonOptions.model` and
`SideQueryTextOptions.model` to reflect the actual fallback chain
(`getFastModelForSideQuery` → `getFastModel` → `getModel` →
`DEFAULT_QWEN_MODEL`) introduced in this PR.

packages/cli/src/services/DynamicCommandLocalizationService.ts

@@ -375,7 +375,6 @@ export class DynamicCommandLocalizationService {
     items: TranslationItem[],
     signal: AbortSignal,
   ): Promise<Map<string, string>> {
-    const model = config.getFastModel() ?? config.getModel();
     const targetLanguageName = getLanguageNameForTranslationTarget(language);
     const translations = new Map<string, string>();
 
@@ -391,7 +390,6 @@ export class DynamicCommandLocalizationService {
       try {
         response = await runSideQuery<Record<string, unknown>>(config, {
           purpose: 'dynamic-command-localization',
-          model,
           contents: [{ role: 'user', parts: [{ text: prompt }] }],
           schema: {
             type: 'object',

packages/cli/src/ui/commands/modelCommand.test.ts

@@ -395,7 +395,108 @@ describe('modelCommand', () => {
     });
   });
 
-  it('should reject unavailable fast models for the current auth type', async () => {
+  it('should set fast models configured under another auth type', async () => {
+    const setValue = vi.fn();
+    const setFastModel = vi.fn();
+    mockContext = createMockCommandContext({
+      invocation: {
+        raw: '/model --fast deepseek-v4-flash',
+        name: 'model',
+        args: '--fast deepseek-v4-flash',
+      },
+      services: {
+        config: {
+          getContentGeneratorConfig: vi.fn().mockReturnValue({
+            model: 'claude-opus-4-7',
+            authType: AuthType.USE_ANTHROPIC,
+          }),
+          getAllConfiguredModels: vi.fn().mockReturnValue([
+            {
+              id: 'deepseek-v4-flash',
+              label: 'deepseek-v4-flash',
+              authType: AuthType.USE_OPENAI,
+            },
+            {
+              id: 'claude-opus-4-7',
+              label: 'claude-opus-4-7',
+              authType: AuthType.USE_ANTHROPIC,
+            },
+          ]),
+          setFastModel,
+        },
+        settings: createMockSettings(setValue),
+      },
+    });
+
+    const result = await modelCommand.action!(
+      mockContext,
+      '--fast deepseek-v4-flash',
+    );
+
+    expect(setValue).toHaveBeenCalledWith(
+      expect.any(String),
+      'fastModel',
+      'deepseek-v4-flash',
+    );
+    expect(setFastModel).toHaveBeenCalledWith('deepseek-v4-flash');
+    expect(result).toEqual({
+      type: 'message',
+      messageType: 'info',
+      content: 'Fast Model: deepseek-v4-flash',
+    });
+  });
+
+  it('should set authType-qualified fast model selectors', async () => {
+    const setValue = vi.fn();
+    const setFastModel = vi.fn();
+    mockContext = createMockCommandContext({
+      invocation: {
+        raw: '/model --fast openai:deepseek-v4-flash',
+        name: 'model',
+        args: '--fast openai:deepseek-v4-flash',
+      },
+      services: {
+        config: {
+          getContentGeneratorConfig: vi.fn().mockReturnValue({
+            model: 'claude-opus-4-7',
+            authType: AuthType.USE_ANTHROPIC,
+          }),
+          getAvailableModelsForAuthType: vi.fn((authType: AuthType) =>
+            authType === AuthType.USE_OPENAI
+              ? [
+                  {
+                    id: 'deepseek-v4-flash',
+                    label: 'deepseek-v4-flash',
+                    authType: AuthType.USE_OPENAI,
+                  },
+                ]
+              : [],
+          ),
+          setFastModel,
+        },
+        settings: createMockSettings(setValue),
+      },
+    });
+
+    const result = await modelCommand.action!(
+      mockContext,
+      '--fast openai:deepseek-v4-flash',
+    );
+
+    expect(setValue).toHaveBeenCalledWith(
+      expect.any(String),
+      'fastModel',
+      'openai:deepseek-v4-flash',
+    );
+    expect(setFastModel).toHaveBeenCalledWith('openai:deepseek-v4-flash');
+    expect(result).toEqual({
+      type: 'message',
+      messageType: 'info',
+      content: 'Fast Model: openai:deepseek-v4-flash',
+    });
+  });
+
+  it('should reject unavailable fast models across all auth types', async () => {
     const setValue = vi.fn();
     const setFastModel = vi.fn();
     mockContext = createMockCommandContext({
@@ -410,9 +511,13 @@ describe('modelCommand', () => {
             model: 'qwen-plus',
             authType: AuthType.USE_OPENAI,
           }),
-          getAvailableModelsForAuthType: vi
-            .fn()
-            .mockReturnValue([{ id: 'qwen-turbo', label: 'Qwen Turbo' }]),
+          getAllConfiguredModels: vi.fn().mockReturnValue([
+            {
+              id: 'qwen-turbo',
+              label: 'Qwen Turbo',
+              authType: AuthType.USE_OPENAI,
+            },
+          ]),
           setFastModel,
         },
         settings: createMockSettings(setValue),
@@ -430,8 +535,8 @@ describe('modelCommand', () => {
       type: 'message',
       messageType: 'error',
       content:
-        "Fast model 'missing-model' is not available for auth type 'openai'.\n" +
-        "Available models for 'openai': qwen-turbo.\n" +
+        "Fast model 'missing-model' is not configured for any auth type.\n" +
+        'Configured models: qwen-turbo.\n' +
         'Configure models in settings.modelProviders or run /model to select an available model.',
     });
   });

packages/cli/src/ui/commands/modelCommand.ts

@@ -17,6 +17,7 @@ import {
   AuthType,
   type AvailableModel,
   type Config,
+  resolveModelId,
 } from '@qwen-code/qwen-code-core';
 import type { LoadedSettings } from '../../config/settings.js';
 import { parseAcpModelOption } from '../../utils/acpModelUtils.js';
@@ -77,6 +78,25 @@ function formatUnavailableModelMessage(
   );
 }
 
+function formatUnavailableFastModelMessage(
+  modelName: string,
+  availableModels: AvailableModel[],
+): string {
+  const availableModelIds = Array.from(
+    new Set(availableModels.map((model) => model.id)),
+  );
+  const availableModelsLine =
+    availableModelIds.length === 0
+      ? 'No models are configured.'
+      : `Configured models: ${availableModelIds.join(', ')}.`;
+
+  return (
+    `Fast model '${modelName}' is not configured for any auth type.\n` +
+    `${availableModelsLine}\n` +
+    'Configure models in settings.modelProviders or run /model to select an available model.'
+  );
+}
+
 // Get an array of the available model IDs as strings
 function getAvailableModelIds(context: CommandContext) {
   const { services } = context;
@@ -174,17 +194,36 @@ export const modelCommand: SlashCommand = {
         };
       }
 
-      const availableModels = config.getAvailableModelsForAuthType(authType);
-      if (!availableModels.some((model) => model.id === modelName)) {
+      const selector = (() => {
+        try {
+          return resolveModelId(modelName);
+        } catch {
+          return undefined;
+        }
+      })();
+      if (!selector) {
         return {
           type: 'message',
           messageType: 'error',
-          content: formatUnavailableModelMessage(
-            'Fast model',
-            modelName,
-            authType,
-            availableModels,
-          ),
+          content: formatUnavailableFastModelMessage(modelName, []),
+        };
+      }
+
+      const availableModels = selector.authType
+        ? config.getAvailableModelsForAuthType(selector.authType)
+        : config.getAllConfiguredModels();
+      if (!availableModels.some((model) => model.id === selector.modelId)) {
+        return {
+          type: 'message',
+          messageType: 'error',
+          content: selector.authType
+            ? formatUnavailableModelMessage(
+                'Fast model',
+                selector.modelId,
+                selector.authType,
+                availableModels,
+              )
+            : formatUnavailableFastModelMessage(modelName, availableModels),
         };
       }
 

packages/cli/src/ui/commands/renameCommand.ts

@@ -213,7 +213,9 @@ export const renameCommand: SlashCommand = {
       // the main model here because `--auto` is a deliberate opt-in to the
       // sentence-case fast-model flow, and surprising a user with a main-
       // model call would defeat the purpose.
-      if (!config.getFastModel()) {
+      const fastModel =
+        config.getFastModelForSideQuery?.() ?? config.getFastModel();
+      if (!fastModel) {
         return {
           type: 'message',
           messageType: 'error',

packages/cli/src/ui/components/ModelDialog.test.tsx

@@ -288,6 +288,108 @@ describe('<ModelDialog />', () => {
     expect(props.onClose).toHaveBeenCalledTimes(1);
   });
 
+  it('stores authType-qualified selectors in fast model mode', async () => {
+    const setFastModel = vi.fn();
+    const { props, mockSettings } = renderComponent({ isFastModelMode: true }, {
+      getAuthType: vi.fn(() => AuthType.USE_ANTHROPIC),
+      getModel: vi.fn(() => 'claude-opus-4-7'),
+      getAllConfiguredModels: vi.fn(() => [
+        {
+          id: 'deepseek-v4-flash',
+          label: 'deepseek-v4-flash',
+          authType: AuthType.USE_OPENAI,
+        },
+        {
+          id: 'claude-opus-4-7',
+          label: 'claude-opus-4-7',
+          authType: AuthType.USE_ANTHROPIC,
+        },
+      ]),
+      getContentGeneratorConfig: vi.fn(() => ({
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'claude-opus-4-7',
+      })),
+      setFastModel,
+    } as unknown as Partial<Config>);
+
+    const childOnSelect = mockedSelect.mock.calls[0][0].onSelect;
+    await childOnSelect(`${AuthType.USE_OPENAI}::deepseek-v4-flash`);
+
+    expect(mockSettings.setValue).toHaveBeenCalledWith(
+      SettingScope.User,
+      'fastModel',
+      'openai:deepseek-v4-flash',
+    );
+    expect(setFastModel).toHaveBeenCalledWith('openai:deepseek-v4-flash');
+    expect(props.onClose).toHaveBeenCalledTimes(1);
+  });
+
+  it('highlights the cross-auth row for a bare fast-model setting', () => {
+    // `/model --fast deepseek-v4-flash` validates across all providers and
+    // persists the bare model id. When the dialog re-opens, it must locate
+    // the right row even though the setting carries no authType prefix —
+    // otherwise the highlight falls back to the current auth's first row
+    // and Enter would silently overwrite the setting.
+    const mockSettings = {
+      isTrusted: true,
+      user: { settings: {} },
+      workspace: { settings: {} },
+      merged: { fastModel: 'deepseek-v4-flash' },
+      setValue: vi.fn(),
+    } as unknown as LoadedSettings;
+
+    const allModels = [
+      {
+        id: 'claude-opus-4-7',
+        label: 'claude-opus-4-7',
+        description: '',
+        authType: AuthType.USE_ANTHROPIC,
+      },
+      {
+        id: 'deepseek-v4-flash',
+        label: 'deepseek-v4-flash',
+        description: '',
+        authType: AuthType.USE_OPENAI,
+      },
+    ];
+
+    render(
+      <SettingsContext.Provider value={mockSettings}>
+        <ConfigContext.Provider
+          value={
+            {
+              getModel: vi.fn(() => 'claude-opus-4-7'),
+              getAuthType: vi.fn(() => AuthType.USE_ANTHROPIC),
+              getAllConfiguredModels: vi.fn(() => allModels),
+              getContentGeneratorConfig: vi.fn(() => ({
+                authType: AuthType.USE_ANTHROPIC,
+                model: 'claude-opus-4-7',
+              })),
+              getModelsConfig: vi.fn(() => ({
+                getGenerationConfig: vi.fn(() => ({ baseUrl: undefined })),
+              })),
+              getActiveRuntimeModelSnapshot: vi.fn(() => undefined),
+              getUsageStatisticsEnabled: vi.fn(() => false),
+              getSessionId: vi.fn(() => 'session'),
+              getDebugMode: vi.fn(() => false),
+              getUseModelRouter: vi.fn(() => false),
+              getProxy: vi.fn(() => undefined),
+            } as unknown as Config
+          }
+        >
+          <ModelDialog onClose={vi.fn()} isFastModelMode={true} />
+        </ConfigContext.Provider>
+      </SettingsContext.Provider>,
+    );
+
+    const items = mockedSelect.mock.calls[0][0].items;
+    const deepseekIndex = items.findIndex((item) =>
+      String(item.value).includes('deepseek-v4-flash'),
+    );
+    expect(deepseekIndex).toBeGreaterThanOrEqual(0);
+    expect(mockedSelect.mock.calls[0][0].initialIndex).toBe(deepseekIndex);
+  });
+
   it('blocks switching to qwen-oauth from another authType (discontinued)', async () => {
     const switchModel = vi.fn().mockResolvedValue(undefined);
     const getAuthType = vi.fn(() => AuthType.USE_OPENAI);

packages/cli/src/ui/components/ModelDialog.tsx

@@ -12,6 +12,7 @@ import {
   ModelSlashCommandEvent,
   logModelSlashCommand,
   MAINLINE_CODER_MODEL,
+  resolveModelId,
   type AvailableModel as CoreAvailableModel,
   type ContentGeneratorConfig,
   type InputModalities,
@@ -303,9 +304,17 @@ export function ModelDialog({
 
   // In fast model mode, default to the currently configured fast model
   const fastModelSetting = settings?.merged?.fastModel as string | undefined;
+  const parsedFastModelSetting = useMemo(() => {
+    if (!isFastModelMode) return undefined;
+    try {
+      return resolveModelId(fastModelSetting);
+    } catch {
+      return undefined;
+    }
+  }, [fastModelSetting, isFastModelMode]);
   const preferredModelId =
-    isFastModelMode && fastModelSetting
-      ? fastModelSetting
+    isFastModelMode && parsedFastModelSetting
+      ? parsedFastModelSetting.modelId
       : config?.getModel() || MAINLINE_CODER_MODEL;
   // Check if current model is a runtime model
   // Runtime snapshot ID is already in $runtime|${authType}|${modelId} format
@@ -315,11 +324,35 @@ export function ModelDialog({
   const currentBaseUrl = config
     ?.getModelsConfig()
     .getGenerationConfig()?.baseUrl;
+  // When `/model --fast <bare-id>` validated the model across all providers,
+  // the setting persists as a bare model ID (no authType prefix) so that
+  // runtime cross-auth lookups still work. Highlight the row that owns it
+  // regardless of which provider that turns out to be — otherwise the
+  // dialog would default to the current auth's first row and Enter would
+  // silently overwrite the user's fast-model setting.
+  const preferredFastModelEntry =
+    isFastModelMode && parsedFastModelSetting
+      ? parsedFastModelSetting.authType
+        ? availableModelEntries.find(
+            ({ authType: t2, model }) =>
+              t2 === parsedFastModelSetting.authType &&
+              model.id === parsedFastModelSetting.modelId,
+          )
+        : availableModelEntries.find(
+            ({ model }) => model.id === parsedFastModelSetting.modelId,
+          )
+      : undefined;
   const preferredKey = activeRuntimeSnapshot
     ? activeRuntimeSnapshot.id
-    : authType
-      ? buildModelSelectionKey(authType, preferredModelId, currentBaseUrl)
-      : '';
+    : preferredFastModelEntry
+      ? buildModelSelectionKey(
+          preferredFastModelEntry.authType,
+          preferredFastModelEntry.model.id,
+          preferredFastModelEntry.model.baseUrl,
+        )
+      : authType
+        ? buildModelSelectionKey(authType, preferredModelId, currentBaseUrl)
+        : '';
 
   useKeypress(
     (key) => {
@@ -358,27 +391,28 @@ export function ModelDialog({
     async (selected: string) => {
       setErrorMessage(null);
 
-      // Fast model mode: save the model ID only (baseUrl is intentionally
-      // discarded — getFastModel resolves via the first registry match).
+      // Fast model mode: save authType:modelId so duplicate model ids across
+      // providers remain unambiguous. baseUrl is intentionally discarded.
       if (isFastModelMode) {
-        let modelId: string;
+        let fastModel: string;
         if (selected.includes('::')) {
           const parsed = parseModelSelectionKey(selected);
-          modelId = parsed.modelId;
+          fastModel = `${parsed.authType}:${parsed.modelId}`;
         } else if (selected.startsWith('$runtime|')) {
           const parts = selected.split('|');
-          modelId = parts[2] ?? selected;
+          fastModel =
+            parts[1] && parts[2] ? `${parts[1]}:${parts[2]}` : selected;
         } else {
-          modelId = selected;
+          fastModel = selected;
         }
         const scope = getPersistScopeForModelSelection(settings);
-        settings.setValue(scope, 'fastModel', modelId);
+        settings.setValue(scope, 'fastModel', fastModel);
         // Sync the runtime Config so forked agents pick up the change immediately.
-        config?.setFastModel(modelId);
+        config?.setFastModel(fastModel);
         uiState?.historyManager.addItem(
           {
             type: 'success',
-            text: `${t('Fast Model')}: ${modelId}`,
+            text: `${t('Fast Model')}: ${fastModel}`,
           },
           Date.now(),
         );

packages/core/src/config/config.test.ts

@@ -760,6 +760,190 @@ describe('Server Config (config.ts)', () => {
   });
 
   describe('model switching with different credentials (OpenAI)', () => {
+    it('keeps getFastModel current-auth-only for direct runtime callers', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'claude-opus-4-7',
+        fastModel: 'deepseek-v4-flash',
+        modelProvidersConfig: {
+          [AuthType.USE_OPENAI]: [
+            {
+              id: 'deepseek-v4-flash',
+              name: 'deepseek-v4-flash',
+              baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+              envKey: 'DASHSCOPE_API_KEY',
+            },
+          ],
+          [AuthType.USE_ANTHROPIC]: [
+            {
+              id: 'claude-opus-4-7',
+              name: 'claude-opus-4-7',
+              baseUrl: 'https://idealab.alibaba-inc.com/api/anthropic',
+              envKey: 'IDEALAB_OPUS_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBeUndefined();
+      expect(config.getFastModelForSideQuery()).toBe('deepseek-v4-flash');
+    });
+
+    it('returns an authType-qualified fast model selector for side queries', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'shared-model',
+        fastModel: 'openai:shared-model',
+        modelProvidersConfig: {
+          [AuthType.USE_OPENAI]: [
+            {
+              id: 'shared-model',
+              name: 'OpenAI shared model',
+              baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+              envKey: 'DASHSCOPE_API_KEY',
+            },
+          ],
+          [AuthType.USE_ANTHROPIC]: [
+            {
+              id: 'shared-model',
+              name: 'Anthropic shared model',
+              baseUrl: 'https://idealab.alibaba-inc.com/api/anthropic',
+              envKey: 'IDEALAB_OPUS_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBeUndefined();
+      expect(config.getFastModelForSideQuery()).toBe('openai:shared-model');
+    });
+
+    it('returns a bare fast model for getFastModel when authType-qualified selector matches the current auth type', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_OPENAI,
+        model: 'gpt-4',
+        fastModel: 'openai:deepseek-v4-flash',
+        modelProvidersConfig: {
+          [AuthType.USE_OPENAI]: [
+            {
+              id: 'deepseek-v4-flash',
+              name: 'deepseek-v4-flash',
+              baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+              envKey: 'DASHSCOPE_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBe('deepseek-v4-flash');
+      expect(config.getFastModelForSideQuery()).toBe(
+        'openai:deepseek-v4-flash',
+      );
+    });
+
+    it('accepts runtime fast models for authType-qualified selectors', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_OPENAI,
+        model: 'runtime-fast-model',
+        fastModel: 'openai:runtime-fast-model',
+        generationConfig: {
+          apiKey: 'sk-runtime-key',
+          baseUrl: 'https://runtime.example.com/v1',
+        },
+        generationConfigSources: {
+          model: { kind: 'programmatic', detail: 'test' },
+          apiKey: { kind: 'programmatic', detail: 'test' },
+          baseUrl: { kind: 'programmatic', detail: 'test' },
+        },
+        modelProvidersConfig: {
+          [AuthType.USE_OPENAI]: [
+            {
+              id: 'registry-model',
+              name: 'Registry Model',
+              baseUrl: 'https://api.openai.com/v1',
+              envKey: 'OPENAI_API_KEY',
+            },
+          ],
+        },
+      });
+      config.getModelsConfig().detectAndCaptureRuntimeModel();
+
+      expect(config.getFastModel()).toBe('runtime-fast-model');
+      expect(config.getFastModelForSideQuery()).toBe(
+        'openai:runtime-fast-model',
+      );
+    });
+
+    it('returns undefined when the fast model is not configured for any auth type', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'claude-opus-4-7',
+        fastModel: 'missing-fast-model',
+        modelProvidersConfig: {
+          [AuthType.USE_ANTHROPIC]: [
+            {
+              id: 'claude-opus-4-7',
+              name: 'claude-opus-4-7',
+              baseUrl: 'https://idealab.alibaba-inc.com/api/anthropic',
+              envKey: 'IDEALAB_OPUS_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBeUndefined();
+      expect(config.getFastModelForSideQuery()).toBeUndefined();
+    });
+
+    it('returns undefined when the fast model selector is malformed', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'claude-opus-4-7',
+        fastModel: 'openai:',
+        modelProvidersConfig: {
+          [AuthType.USE_OPENAI]: [
+            {
+              id: 'deepseek-v4-flash',
+              name: 'deepseek-v4-flash',
+              baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+              envKey: 'DASHSCOPE_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBeUndefined();
+      expect(config.getFastModelForSideQuery()).toBeUndefined();
+    });
+
+    it('returns undefined when fastModel points back to the fast selector', () => {
+      const config = new Config({
+        ...baseParams,
+        authType: AuthType.USE_ANTHROPIC,
+        model: 'claude-opus-4-7',
+        fastModel: 'fast',
+        modelProvidersConfig: {
+          [AuthType.USE_ANTHROPIC]: [
+            {
+              id: 'claude-opus-4-7',
+              name: 'claude-opus-4-7',
+              baseUrl: 'https://idealab.alibaba-inc.com/api/anthropic',
+              envKey: 'IDEALAB_OPUS_API_KEY',
+            },
+          ],
+        },
+      });
+
+      expect(config.getFastModel()).toBeUndefined();
+      expect(config.getFastModelForSideQuery()).toBeUndefined();
+    });
+
     it('should refresh auth when switching to model with different envKey', async () => {
       // This test verifies the fix for switching between modelProvider models
       // with different envKeys (e.g., deepseek-chat with DEEPSEEK_API_KEY)

packages/core/src/config/config.ts

@@ -154,6 +154,7 @@ import {
   type AvailableModel,
   type RuntimeModelSnapshot,
 } from '../models/index.js';
+import { resolveModelId } from '../utils/modelId.js';
 import type { ClaudeMarketplaceConfig } from '../extension/claude-converter.js';
 
 // Re-export types
@@ -1832,20 +1833,57 @@ export class Config {
   }
 
   /**
-   * Returns the fast model if one is configured and valid for the current auth type,
-   * otherwise returns undefined. Background agents (memory extraction, dream, /btw)
-   * use this as a cheaper alternative to the main session model.
+   * Returns the fast model if one is configured and valid for the current auth
+   * type, otherwise returns undefined. Direct runtime paths use this as a
+   * cheaper alternative to the main session model, so it intentionally stays
+   * current-auth-only.
    */
   getFastModel(): string | undefined {
-    if (!this.fastModel) return undefined;
-    const authType = this.contentGeneratorConfig?.authType;
+    const authType =
+      this.contentGeneratorConfig?.authType ??
+      this.modelsConfig.getCurrentAuthType();
     if (!authType) return undefined;
-    const available = this.getAvailableModelsForAuthType(authType);
-    return available.some((m) => m.id === this.fastModel)
-      ? this.fastModel
+    const selector = this.resolveFastModelSelector();
+    if (!selector) return undefined;
+    if (selector.authType && selector.authType !== authType) return undefined;
+
+    const available = this.getAllConfiguredModels([authType]);
+    return available.some((m) => m.id === selector.modelId)
+      ? selector.modelId
       : undefined;
   }
 
+  /**
+   * Returns the fast model for side-query paths. Unlike {@link getFastModel},
+   * this can return an authType-qualified selector because BaseLlmClient can
+   * route a single request through a provider different from the main session.
+   */
+  getFastModelForSideQuery(): string | undefined {
+    const selector = this.resolveFastModelSelector();
+    if (!selector) return undefined;
+
+    if (selector.authType) {
+      const available = this.getAllConfiguredModels([selector.authType]);
+      return available.some((m) => m.id === selector.modelId)
+        ? `${selector.authType}:${selector.modelId}`
+        : undefined;
+    }
+
+    const available = this.getAllConfiguredModels();
+    return available.some((m) => m.id === selector.modelId)
+      ? selector.modelId
+      : undefined;
+  }
+
+  private resolveFastModelSelector() {
+    if (!this.fastModel) return undefined;
+    try {
+      return resolveModelId(this.fastModel);
+    } catch {
+      return undefined;
+    }
+  }
+
   /**
    * Update the fast model at runtime (e.g., when the user runs `/model --fast <model>`).
    * Pass undefined or an empty string to clear the fast model override.

packages/core/src/core/baseLlmClient.test.ts

@@ -501,6 +501,8 @@ describe('BaseLlmClient', () => {
           .mockReturnValue({ authType: AuthType.QWEN_OAUTH }),
         getEmbeddingModel: vi.fn().mockReturnValue('test-embedding-model'),
         getModel: vi.fn().mockReturnValue('main-model'),
+        getFastModel: vi.fn().mockReturnValue(undefined),
+        getFastModelForSideQuery: vi.fn().mockReturnValue(undefined),
         getModelsConfig: vi.fn().mockReturnValue({ getResolvedModel }),
       } as unknown as Mocked<Config>;
     });
@@ -635,6 +637,95 @@ describe('BaseLlmClient', () => {
       );
     });
 
+    it('generateJson accepts authType-qualified selectors and sends the bare model id', async () => {
+      getResolvedModel.mockImplementation((authType: string, model: string) => {
+        if (authType === AuthType.USE_OPENAI && model === 'shared-model') {
+          return {
+            id: 'shared-model',
+            authType: AuthType.USE_OPENAI,
+            envKey: 'OPENAI_API_KEY',
+          };
+        }
+        return undefined;
+      });
+      fastGenerateContent.mockResolvedValue(
+        createMockResponseWithFunctionCall({ ok: true }),
+      );
+      vi.mocked(getFunctionCalls).mockReturnValue([
+        { name: 'respond_in_schema', args: { ok: true } },
+      ]);
+
+      const c = new BaseLlmClient(mockContentGenerator, crossProviderConfig);
+
+      await c.generateJson({
+        contents: [{ role: 'user', parts: [{ text: 'go' }] }],
+        schema: { type: 'object' },
+        model: 'openai:shared-model',
+        abortSignal: new AbortController().signal,
+        promptId: 'test',
+      });
+
+      expect(getResolvedModel).toHaveBeenCalledWith(
+        AuthType.USE_OPENAI,
+        'shared-model',
+      );
+      expect(mockBuildAgentContentGeneratorConfig).toHaveBeenCalledWith(
+        crossProviderConfig,
+        'shared-model',
+        expect.objectContaining({ authType: AuthType.USE_OPENAI }),
+      );
+      expect(fastGenerateContent).toHaveBeenCalledWith(
+        expect.objectContaining({ model: 'shared-model' }),
+        'test',
+      );
+    });
+
+    it('generateJson resolves fast selectors through the configured fast model', async () => {
+      crossProviderConfig.getFastModelForSideQuery.mockReturnValue(
+        'openai:shared-model',
+      );
+      getResolvedModel.mockImplementation((authType: string, model: string) => {
+        if (authType === AuthType.USE_OPENAI && model === 'shared-model') {
+          return {
+            id: 'shared-model',
+            authType: AuthType.USE_OPENAI,
+            envKey: 'OPENAI_API_KEY',
+          };
+        }
+        return undefined;
+      });
+      fastGenerateContent.mockResolvedValue(
+        createMockResponseWithFunctionCall({ ok: true }),
+      );
+      vi.mocked(getFunctionCalls).mockReturnValue([
+        { name: 'respond_in_schema', args: { ok: true } },
+      ]);
+
+      const c = new BaseLlmClient(mockContentGenerator, crossProviderConfig);
+
+      await c.generateJson({
+        contents: [{ role: 'user', parts: [{ text: 'go' }] }],
+        schema: { type: 'object' },
+        model: 'fast',
+        abortSignal: new AbortController().signal,
+        promptId: 'test',
+      });
+
+      expect(getResolvedModel).toHaveBeenCalledWith(
+        AuthType.USE_OPENAI,
+        'shared-model',
+      );
+      expect(mockBuildAgentContentGeneratorConfig).toHaveBeenCalledWith(
+        crossProviderConfig,
+        'shared-model',
+        expect.objectContaining({ authType: AuthType.USE_OPENAI }),
+      );
+      expect(fastGenerateContent).toHaveBeenCalledWith(
+        expect.objectContaining({ model: 'shared-model' }),
+        'test',
+      );
+    });
+
     it('generateText routes through the per-model generator and forwards retry authType', async () => {
       getResolvedModel.mockReturnValue({
         authType: AuthType.USE_ANTHROPIC,

packages/core/src/core/baseLlmClient.ts

@@ -19,6 +19,7 @@ import type { ContentGenerator } from './contentGenerator.js';
 import { AuthType, createContentGenerator } from './contentGenerator.js';
 import type { ResolvedModelConfig } from '../models/types.js';
 import { buildAgentContentGeneratorConfig } from '../models/content-generator-config.js';
+import { resolveModelId, type ResolvedModelId } from '../utils/modelId.js';
 import { reportError } from '../utils/errorReporting.js';
 import { getErrorMessage } from '../utils/errors.js';
 import { retryWithBackoff, isUnattendedMode } from '../utils/retry.js';
@@ -41,6 +42,7 @@ const debugLogger = createDebugLogger('BASE_LLM_CLIENT');
 export interface ResolvedGeneratorForModel {
   contentGenerator: ContentGenerator;
   retryAuthType: string | undefined;
+  model: string;
 }
 
 /**
@@ -173,14 +175,17 @@ export class BaseLlmClient {
       },
     ];
 
-    const { contentGenerator, retryAuthType } =
-      await this.resolveForModel(model);
+    const {
+      contentGenerator,
+      retryAuthType,
+      model: requestModel,
+    } = await this.resolveForModel(model);
 
     try {
       const apiCall = () =>
         contentGenerator.generateContent(
           {
-            model,
+            model: requestModel,
             config: {
               ...requestConfig,
               tools,
@@ -265,14 +270,17 @@ export class BaseLlmClient {
       ...(systemInstruction && { systemInstruction }),
     };
 
-    const { contentGenerator, retryAuthType } =
-      await this.resolveForModel(model);
+    const {
+      contentGenerator,
+      retryAuthType,
+      model: requestModel,
+    } = await this.resolveForModel(model);
 
     try {
       const apiCall = () =>
         contentGenerator.generateContent(
           {
-            model,
+            model: requestModel,
             config: requestConfig,
             contents,
           },
@@ -363,23 +371,35 @@ export class BaseLlmClient {
    * or generator creation fails (e.g. tests without full auth setup).
    */
   async resolveForModel(model: string): Promise<ResolvedGeneratorForModel> {
+    const selector = this.resolveModelSelector(model);
+    const requestModel = selector?.modelId ?? this.config.getModel() ?? model;
     const mainModel = this.config.getModel() ?? model;
     const mainAuthType = this.config.getContentGeneratorConfig()?.authType;
 
-    if (model === mainModel) {
+    if (
+      requestModel === mainModel &&
+      (!selector?.authType || selector.authType === mainAuthType)
+    ) {
       return {
         contentGenerator: this.contentGenerator,
         retryAuthType: mainAuthType,
+        model: requestModel,
       };
     }
 
-    const contentGenerator = await this.createContentGeneratorForModel(model);
+    const contentGenerator = await this.createContentGeneratorForModel(
+      model,
+      selector,
+    );
+    const resolvedModel = this.resolveModelAcrossAuthTypes(model, selector);
     const retryAuthType =
-      this.resolveModelAcrossAuthTypes(model)?.authType ??
-      mainAuthType ??
-      AuthType.USE_OPENAI;
+      resolvedModel?.authType ?? mainAuthType ?? AuthType.USE_OPENAI;
 
-    return { contentGenerator, retryAuthType };
+    return {
+      contentGenerator,
+      retryAuthType,
+      model: resolvedModel?.id ?? requestModel,
+    };
   }
 
   /**
@@ -397,9 +417,16 @@ export class BaseLlmClient {
    */
   private resolveModelAcrossAuthTypes(
     model: string,
+    selector: ResolvedModelId | undefined,
   ): ResolvedModelConfig | undefined {
     const modelsConfig = this.config.getModelsConfig?.();
     if (!modelsConfig) return undefined;
+    if (!selector) return undefined;
+    const modelId = selector.modelId;
+
+    if (selector.authType) {
+      return modelsConfig.getResolvedModel(selector.authType, modelId);
+    }
 
     const allAuthTypes: AuthType[] = [
       AuthType.QWEN_OAUTH,
@@ -411,13 +438,13 @@ export class BaseLlmClient {
 
     const mainAuthType = this.config.getContentGeneratorConfig()?.authType;
     if (mainAuthType) {
-      const resolved = modelsConfig.getResolvedModel(mainAuthType, model);
+      const resolved = modelsConfig.getResolvedModel(mainAuthType, modelId);
       if (resolved) return resolved;
     }
 
     for (const authType of allAuthTypes) {
       if (authType === mainAuthType) continue;
-      const resolved = modelsConfig.getResolvedModel(authType, model);
+      const resolved = modelsConfig.getResolvedModel(authType, modelId);
       if (resolved) return resolved;
     }
 
@@ -426,13 +453,17 @@ export class BaseLlmClient {
 
   private async createContentGeneratorForModel(
     model: string,
+    selector: ResolvedModelId | undefined,
   ): Promise<ContentGenerator> {
-    const cached = this.perModelGeneratorCache.get(model);
+    const cacheKey = selector
+      ? `${selector.authType ?? ''}:${selector.modelId}`
+      : model;
+    const cached = this.perModelGeneratorCache.get(cacheKey);
     if (cached) return cached;
 
     const generatorPromise = (async () => {
       try {
-        const resolvedModel = this.resolveModelAcrossAuthTypes(model);
+        const resolvedModel = this.resolveModelAcrossAuthTypes(model, selector);
 
         if (!resolvedModel) {
           debugLogger.warn(
@@ -441,9 +472,10 @@ export class BaseLlmClient {
           return this.contentGenerator;
         }
 
+        const targetModel = resolvedModel.id ?? selector?.modelId ?? model;
         const targetConfig = buildAgentContentGeneratorConfig(
           this.config,
-          model,
+          targetModel,
           {
             authType: resolvedModel.authType,
             apiKey: resolvedModel.envKey
@@ -459,12 +491,22 @@ export class BaseLlmClient {
           `Failed to create content generator for model "${model}", falling back to main generator.`,
           err instanceof Error ? err.message : String(err),
         );
-        this.perModelGeneratorCache.delete(model);
+        this.perModelGeneratorCache.delete(cacheKey);
         return this.contentGenerator;
       }
     })();
 
-    this.perModelGeneratorCache.set(model, generatorPromise);
+    this.perModelGeneratorCache.set(cacheKey, generatorPromise);
     return generatorPromise;
   }
+
+  private resolveModelSelector(model: string): ResolvedModelId | undefined {
+    return resolveModelId(model, {
+      currentModel: this.config.getModel(),
+      currentAuthType: this.config.getContentGeneratorConfig()?.authType,
+      fastModel:
+        this.config.getFastModelForSideQuery?.() ??
+        this.config.getFastModel?.(),
+    });
+  }
 }

packages/core/src/core/client.ts

@@ -1576,16 +1576,18 @@ export class GeminiClient {
           // main model's config. The retry authType is resolved alongside so that
           // provider-specific checks (e.g. QWEN_OAUTH quota detection) reference
           // the target model's provider.
-          const { contentGenerator, retryAuthType } = await this.config
-            .getBaseLlmClient()
-            .resolveForModel(model);
+          const {
+            contentGenerator,
+            retryAuthType,
+            model: requestModel,
+          } = await this.config.getBaseLlmClient().resolveForModel(model);
 
           const apiCall = () => {
-            currentAttemptModel = model;
+            currentAttemptModel = requestModel;
 
             return contentGenerator.generateContent(
               {
-                model,
+                model: requestModel,
                 config: requestConfig,
                 contents,
               },

packages/core/src/index.ts

@@ -286,6 +286,7 @@ export * from './utils/gitUtils.js';
 export * from './utils/ignorePatterns.js';
 export * from './utils/jsonl-utils.js';
 export * from './utils/memoryDiscovery.js';
+export * from './utils/modelId.js';
 export { ConditionalRulesRegistry } from './utils/rulesDiscovery.js';
 export type { RuleFile } from './utils/rulesDiscovery.js';
 export { OpenAILogger, openaiLogger } from './utils/openaiLogger.js';

packages/core/src/memory/relevanceSelector.test.ts

@@ -129,7 +129,7 @@ describe('selectRelevantAutoMemoryDocumentsByModel', () => {
     );
   });
 
-  it('passes the fast model to runSideQuery when configured', async () => {
+  it('lets runSideQuery choose the default side-query model when fast model is configured', async () => {
     vi.mocked(mockConfig.getFastModel).mockReturnValue('fast-flash-model');
     vi.mocked(runSideQuery).mockResolvedValue({
       selected_memories: ['reference.md'],
@@ -146,13 +146,15 @@ describe('selectRelevantAutoMemoryDocumentsByModel', () => {
       mockConfig,
       expect.objectContaining({
         purpose: 'auto-memory-recall',
-        model: 'fast-flash-model',
         config: { temperature: 0 },
       }),
     );
+    expect(
+      'model' in (vi.mocked(runSideQuery).mock.calls[0]![1] as object),
+    ).toBe(false);
   });
 
-  it('passes undefined model when no fast model is configured', async () => {
+  it('lets runSideQuery fall back to its default when no fast model is configured', async () => {
     vi.mocked(mockConfig.getFastModel).mockReturnValue(undefined);
     vi.mocked(runSideQuery).mockResolvedValue({
       selected_memories: ['reference.md'],
@@ -169,10 +171,12 @@ describe('selectRelevantAutoMemoryDocumentsByModel', () => {
       mockConfig,
       expect.objectContaining({
         purpose: 'auto-memory-recall',
-        model: undefined,
         config: { temperature: 0 },
       }),
     );
+    expect(
+      'model' in (vi.mocked(runSideQuery).mock.calls[0]![1] as object),
+    ).toBe(false);
   });
 
   it('throws when selector returns unknown relative paths', async () => {

packages/core/src/memory/relevanceSelector.ts

@@ -95,9 +95,8 @@ export async function selectRelevantAutoMemoryDocumentsByModel(
       ? AbortSignal.any([AbortSignal.timeout(1_000), callerAbortSignal])
       : AbortSignal.timeout(1_000),
 
-    // Use the fast model for this background side-query to reduce latency and
-    // cost. Falls back to the main session model if no fast model is configured.
-    model: config.getFastModel(),
+    // Uses runSideQuery's default side-query model policy: fast model first,
+    // then main session model when no fast model is configured.
     systemInstruction: SELECT_MEMORIES_SYSTEM_PROMPT,
     config: {
       temperature: 0,

packages/core/src/services/chatRecordingService.ts

@@ -928,11 +928,12 @@ export class ChatRecordingService {
     // Headless/one-shot CLI flows (`qwen -p "…"`, cron, CI scripts) run a
     // single prompt and throw the session away. Spending fast-model tokens
     // on a title no one will ever resume is pure waste; skip entirely.
-    // Checked before `getFastModel()` because it's strictly cheaper (a bool
-    // field read vs. a method that looks up available models for the auth
-    // type).
+    // Checked before `getFastModelForSideQuery()` because it's strictly
+    // cheaper (a bool field read vs. a method that looks up available models).
     if (!this.config.isInteractive()) return;
-    if (!this.config.getFastModel()) return;
+    const fastModel =
+      this.config.getFastModelForSideQuery?.() ?? this.config.getFastModel();
+    if (!fastModel) return;
 
     this.autoTitleAttempts++;
     const controller = new AbortController();

packages/core/src/services/sessionRecap.ts

@@ -61,7 +61,10 @@ export async function generateSessionRecap(
     const recentHistory = takeRecentDialog(dialog, RECENT_MESSAGE_WINDOW);
     if (recentHistory.length === 0) return null;
 
-    const model = config.getFastModel() ?? config.getModel();
+    const model =
+      config.getFastModelForSideQuery?.() ??
+      config.getFastModel() ??
+      config.getModel();
 
     const result = await runSideQuery(config, {
       purpose: 'session-recap',
@@ -75,7 +78,6 @@ export async function generateSessionRecap(
         temperature: 0.3,
       },
       abortSignal,
-      model,
       // Recap is best-effort cosmetic — don't burn the default 7 retries.
       maxAttempts: 1,
     });

packages/core/src/services/sessionTitle.ts

@@ -70,8 +70,8 @@ const TRAILING_PAIRED_BRACKETS_RE =
  * command) can surface actionable messages instead of a generic "could not
  * generate".
  *
- * - `no_fast_model`: config.getFastModel() returned undefined. User needs to
- *   configure one via `/model --fast <name>`.
+ * - `no_fast_model`: config.getFastModelForSideQuery() returned undefined.
+ *   User needs to configure one via `/model --fast <name>`.
  * - `no_client`: BaseLlmClient or GeminiClient not yet initialized. Rare,
  *   usually means the session hasn't authenticated yet.
  * - `empty_history`: the conversation has fewer than 2 turns of usable text.
@@ -107,7 +107,7 @@ export async function tryGenerateSessionTitle(
   abortSignal: AbortSignal,
 ): Promise<SessionTitleOutcome> {
   try {
-    const model = config.getFastModel();
+    const model = config.getFastModelForSideQuery?.() ?? config.getFastModel();
     if (!model) return { ok: false, reason: 'no_fast_model' };
 
     const geminiClient = config.getGeminiClient();
@@ -130,7 +130,6 @@ export async function tryGenerateSessionTitle(
 
     const result = await runSideQuery<{ title?: string }>(config, {
       purpose: 'session-title',
-      model,
       systemInstruction: TITLE_SYSTEM_PROMPT,
       schema: TITLE_SCHEMA as unknown as Record<string, unknown>,
       contents: [

packages/core/src/services/toolUseSummary.ts

@@ -89,7 +89,8 @@ export interface GenerateToolUseSummaryParams {
    */
   lastAssistantText?: string;
   /**
-   * Fast model to use. If omitted, falls back to `config.getFastModel()`;
+   * Fast model to use. If omitted, falls back to
+   * `config.getFastModelForSideQuery()`;
    * if that also returns undefined, the call is skipped (returns null).
    * Unlike `sessionRecap`, this does not fall back to the main model —
    * summary generation is a nice-to-have and must not incur main-model cost.
@@ -112,7 +113,10 @@ export async function generateToolUseSummary(
     return null;
   }
 
-  const model = params.model ?? config.getFastModel();
+  const model =
+    params.model ??
+    config.getFastModelForSideQuery?.() ??
+    config.getFastModel();
   if (!model) {
     debugLogger.debug('No fast model configured — skipping summary generation');
     return null;
@@ -151,7 +155,7 @@ export async function generateToolUseSummary(
         temperature: 0.3,
       },
       abortSignal: signal,
-      model,
+      ...(params.model !== undefined ? { model: params.model } : {}),
       // Tool-use labels are best-effort cosmetic; firing once per turn means
       // 7 retries on a transient outage would spike traffic for no benefit.
       maxAttempts: 1,

packages/core/src/utils/modelId.test.ts

@@ -0,0 +1,98 @@
+/**
+ * @license
+ * Copyright 2025 Qwen
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { describe, expect, it } from 'vitest';
+import { AuthType } from '../core/contentGenerator.js';
+import { resolveModelId } from './modelId.js';
+
+describe('resolveModelId', () => {
+  it('returns undefined for omitted models without a current model', () => {
+    expect(resolveModelId(undefined)).toBeUndefined();
+  });
+
+  it('resolves omitted models to the current model when provided', () => {
+    expect(
+      resolveModelId(undefined, {
+        currentModel: 'main-model',
+        currentAuthType: AuthType.USE_ANTHROPIC,
+      }),
+    ).toEqual({
+      authType: AuthType.USE_ANTHROPIC,
+      modelId: 'main-model',
+    });
+  });
+
+  it('resolves explicit inherit to the current model', () => {
+    expect(
+      resolveModelId('inherit', {
+        currentModel: 'main-model',
+        currentAuthType: AuthType.USE_OPENAI,
+      }),
+    ).toEqual({
+      authType: AuthType.USE_OPENAI,
+      modelId: 'main-model',
+    });
+  });
+
+  it('returns undefined for fast when no fast model is available', () => {
+    expect(resolveModelId('fast')).toBeUndefined();
+  });
+
+  it('resolves fast to the configured fast model', () => {
+    expect(resolveModelId('fast', { fastModel: 'fast-model' })).toEqual({
+      modelId: 'fast-model',
+    });
+  });
+
+  it('resolves fast to authType-prefixed configured fast models', () => {
+    expect(resolveModelId('fast', { fastModel: 'openai:fast-model' })).toEqual({
+      authType: AuthType.USE_OPENAI,
+      modelId: 'fast-model',
+    });
+  });
+
+  it('returns undefined for recursive fast selectors', () => {
+    expect(resolveModelId('fast', { fastModel: 'fast' })).toBeUndefined();
+  });
+
+  it('parses bare model IDs to concrete model IDs', () => {
+    expect(resolveModelId('glm-5')).toEqual({
+      modelId: 'glm-5',
+    });
+  });
+
+  it('parses authType-prefixed model IDs', () => {
+    expect(resolveModelId('openai:glm-5')).toEqual({
+      authType: AuthType.USE_OPENAI,
+      modelId: 'glm-5',
+    });
+  });
+
+  it('trims authType-prefixed model IDs', () => {
+    expect(resolveModelId(' openai : glm-5 ')).toEqual({
+      authType: AuthType.USE_OPENAI,
+      modelId: 'glm-5',
+    });
+  });
+
+  it('treats unknown prefix as bare model ID (colon in model ID)', () => {
+    expect(resolveModelId('invalid:glm-5')).toEqual({
+      modelId: 'invalid:glm-5',
+    });
+  });
+
+  it('treats model IDs with colons as bare model IDs', () => {
+    expect(resolveModelId('gpt-4o:online')).toEqual({
+      modelId: 'gpt-4o:online',
+    });
+  });
+
+  it('rejects missing model IDs after valid authType prefixes', () => {
+    expect(() => resolveModelId('openai:')).toThrow(
+      'Model selector must include a model ID after the authType',
+    );
+  });
+});

packages/core/src/utils/modelId.ts

@@ -0,0 +1,122 @@
+/**
+ * @license
+ * Copyright 2025 Qwen
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { AuthType } from '../core/contentGenerator.js';
+
+export interface ResolvedModelId {
+  authType?: AuthType;
+  modelId: string;
+}
+
+export interface ModelIdResolutionContext {
+  currentModel?: string;
+  currentAuthType?: AuthType;
+  fastModel?: string;
+}
+
+type ModelIdSelector =
+  | {
+      kind: 'inherit';
+    }
+  | {
+      kind: 'fast';
+    }
+  | {
+      kind: 'model';
+      authType?: AuthType;
+      modelId: string;
+    };
+
+const AUTH_TYPES = new Set<AuthType>(Object.values(AuthType));
+
+/**
+ * Resolve a model selector to the concrete model ID a caller should use.
+ *
+ * Supported forms:
+ * - omitted / inherit -> use parent conversation model
+ * - fast -> use the configured fastModel
+ * - modelId -> use parent authType with the provided modelId
+ * - authType:modelId -> use explicit authType and modelId
+ */
+export function resolveModelId(
+  model: string | undefined,
+  context: ModelIdResolutionContext = {},
+): ResolvedModelId | undefined {
+  return resolveModelIdSelector(parseModelIdSelector(model), context);
+}
+
+function parseModelIdSelector(model: string | undefined): ModelIdSelector {
+  const trimmed = model?.trim();
+  if (!trimmed || trimmed === 'inherit') {
+    return { kind: 'inherit' };
+  }
+  if (trimmed === 'fast') {
+    return { kind: 'fast' };
+  }
+
+  const colonIndex = trimmed.indexOf(':');
+  if (colonIndex === -1) {
+    return { kind: 'model', modelId: trimmed };
+  }
+
+  const maybeAuthType = trimmed.slice(0, colonIndex).trim();
+  const modelId = trimmed.slice(colonIndex + 1).trim();
+
+  // If the prefix isn't a known AuthType, treat the whole string as a bare
+  // model ID. Model IDs can legitimately contain colons (e.g. gpt-4o:online).
+  if (!AUTH_TYPES.has(maybeAuthType as AuthType)) {
+    return { kind: 'model', modelId: trimmed };
+  }
+
+  if (!modelId) {
+    throw new Error(
+      'Model selector must include a model ID after the authType',
+    );
+  }
+
+  return {
+    kind: 'model',
+    authType: maybeAuthType as AuthType,
+    modelId,
+  };
+}
+
+function resolveModelIdSelector(
+  selector: ModelIdSelector,
+  context: ModelIdResolutionContext,
+): ResolvedModelId | undefined {
+  if (selector.kind === 'model') {
+    return {
+      ...(selector.authType ? { authType: selector.authType } : {}),
+      modelId: selector.modelId,
+    };
+  }
+
+  if (selector.kind === 'inherit') {
+    return context.currentModel
+      ? {
+          ...(context.currentAuthType
+            ? { authType: context.currentAuthType }
+            : {}),
+          modelId: context.currentModel,
+        }
+      : undefined;
+  }
+
+  if (!context.fastModel) {
+    return undefined;
+  }
+
+  const fastSelector = parseModelIdSelector(context.fastModel);
+  if (fastSelector.kind === 'fast') {
+    return undefined;
+  }
+
+  return resolveModelIdSelector(fastSelector, {
+    ...context,
+    fastModel: undefined,
+  });
+}

packages/core/src/utils/sideQuery.ts

@@ -20,9 +20,11 @@ export interface SideQueryJsonOptions<TResponse> {
   abortSignal: AbortSignal;
   /**
    * Override the model used for this query. Defaults to
-   * `config.getFastModel?.() ?? config.getModel()` — side queries run on the
-   * fast model when one is configured. Pass an explicit value to pin to the
-   * main model (e.g. long-form summarization in web-fetch).
+   * `config.getFastModelForSideQuery?.() ?? config.getFastModel?.() ?? config.getModel() ?? DEFAULT_QWEN_MODEL`
+   * — side queries run on the fast model when one is configured, including
+   * fast models registered under a different authType than the main session.
+   * Pass an explicit value to pin to the main model (e.g. long-form
+   * summarization in web-fetch).
    */
   model?: string;
   systemInstruction?: string | Part | Part[] | Content;
@@ -61,9 +63,11 @@ export interface SideQueryTextOptions {
   abortSignal: AbortSignal;
   /**
    * Override the model used for this query. Defaults to
-   * `config.getFastModel?.() ?? config.getModel()` — side queries run on the
-   * fast model when one is configured. Pass an explicit value to pin to the
-   * main model (e.g. long-form summarization in web-fetch).
+   * `config.getFastModelForSideQuery?.() ?? config.getFastModel?.() ?? config.getModel() ?? DEFAULT_QWEN_MODEL`
+   * — side queries run on the fast model when one is configured, including
+   * fast models registered under a different authType than the main session.
+   * Pass an explicit value to pin to the main model (e.g. long-form
+   * summarization in web-fetch).
    */
   model?: string;
   systemInstruction?: string | Part | Part[] | Content;
@@ -101,6 +105,7 @@ function buildDefaultPromptId(purpose?: string): string {
 function resolveDefaultModel(config: Config, override?: string): string {
   return (
     override ??
+    config.getFastModelForSideQuery?.() ??
     config.getFastModel?.() ??
     config.getModel() ??
     DEFAULT_QWEN_MODEL