diff --git a/api/server/controllers/agents/client.js b/api/server/controllers/agents/client.js
index 8edbd28122..8e8a993a5d 100644
--- a/api/server/controllers/agents/client.js
+++ b/api/server/controllers/agents/client.js
@@ -39,7 +39,6 @@ const {
   PermissionTypes,
   isAgentsEndpoint,
   isEphemeralAgentId,
-  bedrockInputSchema,
   removeNullishValues,
 } = require('librechat-data-provider');
 const { spendTokens, spendStructuredTokens } = require('~/models/spendTokens');
@@ -69,17 +68,11 @@ const omitTitleOptions = new Set([
  * @param {Agent} agent
  * @param {string} endpoint
  */
-const payloadParser = ({ req, agent, endpoint }) => {
+const payloadParser = ({ req, endpoint }) => {
   if (isAgentsEndpoint(endpoint)) {
-    return { model: undefined };
-  } else if (endpoint === EModelEndpoint.bedrock) {
-    const parsedValues = bedrockInputSchema.parse(agent.model_parameters);
-    if (parsedValues.thinking == null) {
-      parsedValues.thinking = false;
-    }
-    return parsedValues;
+    return;
   }
-  return req.body.endpointOption.model_parameters;
+  return req.body?.endpointOption?.model_parameters;
 };
 
 function createTokenCounter(encoding) {
@@ -296,14 +289,9 @@ class AgentClient extends BaseClient {
   checkVisionRequest() {}
 
   getSaveOptions() {
-    // TODO:
-    // would need to be override settings; otherwise, model needs to be undefined
-    // model: this.override.model,
-    // instructions: this.override.instructions,
-    // additional_instructions: this.override.additional_instructions,
     let runOptions = {};
     try {
-      runOptions = payloadParser(this.options);
+      runOptions = payloadParser(this.options) ?? {};
     } catch (error) {
       logger.error(
         '[api/server/controllers/agents/client.js #getSaveOptions] Error parsing options',
@@ -314,14 +302,14 @@ class AgentClient extends BaseClient {
     return removeNullishValues(
       Object.assign(
         {
+          spec: this.options.spec,
+          iconURL: this.options.iconURL,
           endpoint: this.options.endpoint,
           agent_id: this.options.agent.id,
           modelLabel: this.options.modelLabel,
-          maxContextTokens: this.options.maxContextTokens,
           resendFiles: this.options.resendFiles,
           imageDetail: this.options.imageDetail,
-          spec: this.options.spec,
-          iconURL: this.options.iconURL,
+          maxContextTokens: this.maxContextTokens,
         },
         // TODO: PARSE OPTIONS BY PROVIDER, MAY CONTAIN SENSITIVE DATA
         runOptions,
diff --git a/api/server/middleware/buildEndpointOption.js b/api/server/middleware/buildEndpointOption.js
index f56d850120..64ed8e7466 100644
--- a/api/server/middleware/buildEndpointOption.js
+++ b/api/server/middleware/buildEndpointOption.js
@@ -5,9 +5,11 @@ const {
   EModelEndpoint,
   isAgentsEndpoint,
   parseCompactConvo,
+  getDefaultParamsEndpoint,
 } = require('librechat-data-provider');
 const azureAssistants = require('~/server/services/Endpoints/azureAssistants');
 const assistants = require('~/server/services/Endpoints/assistants');
+const { getEndpointsConfig } = require('~/server/services/Config');
 const agents = require('~/server/services/Endpoints/agents');
 const { updateFilesUsage } = require('~/models');
 
@@ -19,9 +21,24 @@ const buildFunction = {
 
 async function buildEndpointOption(req, res, next) {
   const { endpoint, endpointType } = req.body;
+
+  let endpointsConfig;
+  try {
+    endpointsConfig = await getEndpointsConfig(req);
+  } catch (error) {
+    logger.error('Error fetching endpoints config in buildEndpointOption', error);
+  }
+
+  const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, endpoint);
+
   let parsedBody;
   try {
-    parsedBody = parseCompactConvo({ endpoint, endpointType, conversation: req.body });
+    parsedBody = parseCompactConvo({
+      endpoint,
+      endpointType,
+      conversation: req.body,
+      defaultParamsEndpoint,
+    });
   } catch (error) {
     logger.error(`Error parsing compact conversation for endpoint ${endpoint}`, error);
     logger.debug({
@@ -55,6 +72,7 @@ async function buildEndpointOption(req, res, next) {
         endpoint,
         endpointType,
         conversation: currentModelSpec.preset,
+        defaultParamsEndpoint,
       });
       if (currentModelSpec.iconURL != null && currentModelSpec.iconURL !== '') {
         parsedBody.iconURL = currentModelSpec.iconURL;
diff --git a/api/server/middleware/buildEndpointOption.spec.js b/api/server/middleware/buildEndpointOption.spec.js
new file mode 100644
index 0000000000..eab5e2666b
--- /dev/null
+++ b/api/server/middleware/buildEndpointOption.spec.js
@@ -0,0 +1,237 @@
+/**
+ * Wrap parseCompactConvo: the REAL function runs, but jest can observe
+ * calls and return values. Must be declared before require('./buildEndpointOption')
+ * so the destructured reference in the middleware captures the wrapper.
+ */
+jest.mock('librechat-data-provider', () => {
+  const actual = jest.requireActual('librechat-data-provider');
+  return {
+    ...actual,
+    parseCompactConvo: jest.fn((...args) => actual.parseCompactConvo(...args)),
+  };
+});
+
+const { EModelEndpoint, parseCompactConvo } = require('librechat-data-provider');
+
+const mockBuildOptions = jest.fn((_endpoint, parsedBody) => ({
+  ...parsedBody,
+  endpoint: _endpoint,
+}));
+
+jest.mock('~/server/services/Endpoints/azureAssistants', () => ({
+  buildOptions: mockBuildOptions,
+}));
+jest.mock('~/server/services/Endpoints/assistants', () => ({
+  buildOptions: mockBuildOptions,
+}));
+jest.mock('~/server/services/Endpoints/agents', () => ({
+  buildOptions: mockBuildOptions,
+}));
+
+jest.mock('~/models', () => ({
+  updateFilesUsage: jest.fn(),
+}));
+
+const mockGetEndpointsConfig = jest.fn();
+jest.mock('~/server/services/Config', () => ({
+  getEndpointsConfig: (...args) => mockGetEndpointsConfig(...args),
+}));
+
+jest.mock('@librechat/api', () => ({
+  handleError: jest.fn(),
+}));
+
+const buildEndpointOption = require('./buildEndpointOption');
+
+const createReq = (body, config = {}) => ({
+  body,
+  config,
+  baseUrl: '/api/chat',
+});
+
+const createRes = () => ({
+  status: jest.fn().mockReturnThis(),
+  json: jest.fn().mockReturnThis(),
+});
+
+describe('buildEndpointOption - defaultParamsEndpoint parsing', () => {
+  beforeEach(() => {
+    jest.clearAllMocks();
+  });
+
+  it('should pass defaultParamsEndpoint to parseCompactConvo and preserve maxOutputTokens', async () => {
+    mockGetEndpointsConfig.mockResolvedValue({
+      AnthropicClaude: {
+        type: EModelEndpoint.custom,
+        customParams: {
+          defaultParamsEndpoint: EModelEndpoint.anthropic,
+        },
+      },
+    });
+
+    const req = createReq(
+      {
+        endpoint: 'AnthropicClaude',
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        topP: 0.9,
+        maxContextTokens: 50000,
+      },
+      { modelSpecs: null },
+    );
+
+    await buildEndpointOption(req, createRes(), jest.fn());
+
+    expect(parseCompactConvo).toHaveBeenCalledWith(
+      expect.objectContaining({
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      }),
+    );
+
+    const parsedResult = parseCompactConvo.mock.results[0].value;
+    expect(parsedResult.maxOutputTokens).toBe(8192);
+    expect(parsedResult.topP).toBe(0.9);
+    expect(parsedResult.temperature).toBe(0.7);
+    expect(parsedResult.maxContextTokens).toBe(50000);
+  });
+
+  it('should strip maxOutputTokens when no defaultParamsEndpoint is configured', async () => {
+    mockGetEndpointsConfig.mockResolvedValue({
+      MyOpenRouter: {
+        type: EModelEndpoint.custom,
+      },
+    });
+
+    const req = createReq(
+      {
+        endpoint: 'MyOpenRouter',
+        endpointType: EModelEndpoint.custom,
+        model: 'gpt-4o',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        max_tokens: 4096,
+      },
+      { modelSpecs: null },
+    );
+
+    await buildEndpointOption(req, createRes(), jest.fn());
+
+    expect(parseCompactConvo).toHaveBeenCalledWith(
+      expect.objectContaining({
+        defaultParamsEndpoint: undefined,
+      }),
+    );
+
+    const parsedResult = parseCompactConvo.mock.results[0].value;
+    expect(parsedResult.maxOutputTokens).toBeUndefined();
+    expect(parsedResult.max_tokens).toBe(4096);
+    expect(parsedResult.temperature).toBe(0.7);
+  });
+
+  it('should strip bedrock region from custom endpoint without defaultParamsEndpoint', async () => {
+    mockGetEndpointsConfig.mockResolvedValue({
+      MyEndpoint: {
+        type: EModelEndpoint.custom,
+      },
+    });
+
+    const req = createReq(
+      {
+        endpoint: 'MyEndpoint',
+        endpointType: EModelEndpoint.custom,
+        model: 'gpt-4o',
+        temperature: 0.7,
+        region: 'us-east-1',
+      },
+      { modelSpecs: null },
+    );
+
+    await buildEndpointOption(req, createRes(), jest.fn());
+
+    const parsedResult = parseCompactConvo.mock.results[0].value;
+    expect(parsedResult.region).toBeUndefined();
+    expect(parsedResult.temperature).toBe(0.7);
+  });
+
+  it('should pass defaultParamsEndpoint when re-parsing enforced model spec', async () => {
+    mockGetEndpointsConfig.mockResolvedValue({
+      AnthropicClaude: {
+        type: EModelEndpoint.custom,
+        customParams: {
+          defaultParamsEndpoint: EModelEndpoint.anthropic,
+        },
+      },
+    });
+
+    const modelSpec = {
+      name: 'claude-opus-4.5',
+      preset: {
+        endpoint: 'AnthropicClaude',
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        maxContextTokens: 50000,
+      },
+    };
+
+    const req = createReq(
+      {
+        endpoint: 'AnthropicClaude',
+        endpointType: EModelEndpoint.custom,
+        spec: 'claude-opus-4.5',
+        model: 'anthropic/claude-opus-4.5',
+      },
+      {
+        modelSpecs: {
+          enforce: true,
+          list: [modelSpec],
+        },
+      },
+    );
+
+    await buildEndpointOption(req, createRes(), jest.fn());
+
+    const enforcedCall = parseCompactConvo.mock.calls[1];
+    expect(enforcedCall[0]).toEqual(
+      expect.objectContaining({
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      }),
+    );
+
+    const enforcedResult = parseCompactConvo.mock.results[1].value;
+    expect(enforcedResult.maxOutputTokens).toBe(8192);
+    expect(enforcedResult.temperature).toBe(0.7);
+    expect(enforcedResult.maxContextTokens).toBe(50000);
+  });
+
+  it('should fall back to OpenAI schema when getEndpointsConfig fails', async () => {
+    mockGetEndpointsConfig.mockRejectedValue(new Error('Config unavailable'));
+
+    const req = createReq(
+      {
+        endpoint: 'AnthropicClaude',
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        max_tokens: 4096,
+      },
+      { modelSpecs: null },
+    );
+
+    await buildEndpointOption(req, createRes(), jest.fn());
+
+    expect(parseCompactConvo).toHaveBeenCalledWith(
+      expect.objectContaining({
+        defaultParamsEndpoint: undefined,
+      }),
+    );
+
+    const parsedResult = parseCompactConvo.mock.results[0].value;
+    expect(parsedResult.maxOutputTokens).toBeUndefined();
+    expect(parsedResult.max_tokens).toBe(4096);
+  });
+});
diff --git a/client/src/hooks/Chat/useAddedResponse.ts b/client/src/hooks/Chat/useAddedResponse.ts
index c01cef0c69..fe35e4e56e 100644
--- a/client/src/hooks/Chat/useAddedResponse.ts
+++ b/client/src/hooks/Chat/useAddedResponse.ts
@@ -1,7 +1,12 @@
 import { useCallback } from 'react';
 import { useRecoilValue } from 'recoil';
 import { useGetModelsQuery } from 'librechat-data-provider/react-query';
-import { getEndpointField, LocalStorageKeys, isAssistantsEndpoint } from 'librechat-data-provider';
+import {
+  getEndpointField,
+  LocalStorageKeys,
+  isAssistantsEndpoint,
+  getDefaultParamsEndpoint,
+} from 'librechat-data-provider';
 import type { TEndpointsConfig, EModelEndpoint, TConversation } from 'librechat-data-provider';
 import type { AssistantListItem, NewConversationParams } from '~/common';
 import useAssistantListMap from '~/hooks/Assistants/useAssistantListMap';
@@ -84,11 +89,13 @@ export default function useAddedResponse() {
       }
 
       const models = modelsConfig?.[defaultEndpoint ?? ''] ?? [];
+      const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, defaultEndpoint);
       newConversation = buildDefaultConvo({
         conversation: newConversation,
         lastConversationSetup: preset as TConversation,
         endpoint: defaultEndpoint ?? ('' as EModelEndpoint),
         models,
+        defaultParamsEndpoint,
       });
 
       if (preset?.title != null && preset.title !== '') {
diff --git a/client/src/hooks/Chat/useChatFunctions.ts b/client/src/hooks/Chat/useChatFunctions.ts
index 8479d8eaac..30465fc2e5 100644
--- a/client/src/hooks/Chat/useChatFunctions.ts
+++ b/client/src/hooks/Chat/useChatFunctions.ts
@@ -13,6 +13,7 @@ import {
   parseCompactConvo,
   replaceSpecialVars,
   isAssistantsEndpoint,
+  getDefaultParamsEndpoint,
 } from 'librechat-data-provider';
 import type {
   TMessage,
@@ -173,12 +174,14 @@ export default function useChatFunctions({
     const startupConfig = queryClient.getQueryData<TStartupConfig>([QueryKeys.startupConfig]);
     const endpointType = getEndpointField(endpointsConfig, endpoint, 'type');
     const iconURL = conversation?.iconURL;
+    const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, endpoint);
 
     /** This becomes part of the `endpointOption` */
     const convo = parseCompactConvo({
       endpoint: endpoint as EndpointSchemaKey,
       endpointType: endpointType as EndpointSchemaKey,
       conversation: conversation ?? {},
+      defaultParamsEndpoint,
     });
 
     const { modelDisplayLabel } = endpointsConfig?.[endpoint ?? ''] ?? {};
diff --git a/client/src/hooks/Conversations/useDefaultConvo.ts b/client/src/hooks/Conversations/useDefaultConvo.ts
index bfca39d3e0..67a40ce64e 100644
--- a/client/src/hooks/Conversations/useDefaultConvo.ts
+++ b/client/src/hooks/Conversations/useDefaultConvo.ts
@@ -1,5 +1,5 @@
-import { excludedKeys } from 'librechat-data-provider';
 import { useGetModelsQuery } from 'librechat-data-provider/react-query';
+import { excludedKeys, getDefaultParamsEndpoint } from 'librechat-data-provider';
 import type {
   TEndpointsConfig,
   TModelsConfig,
@@ -47,11 +47,14 @@ const useDefaultConvo = () => {
       }
     }
 
+    const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, endpoint);
+
     const defaultConvo = buildDefaultConvo({
       conversation: conversation as TConversation,
       endpoint,
       lastConversationSetup: preset as TConversation,
       models,
+      defaultParamsEndpoint,
     });
 
     if (!cleanOutput) {
diff --git a/client/src/hooks/Conversations/useGenerateConvo.ts b/client/src/hooks/Conversations/useGenerateConvo.ts
index d96f60e05d..abe3215753 100644
--- a/client/src/hooks/Conversations/useGenerateConvo.ts
+++ b/client/src/hooks/Conversations/useGenerateConvo.ts
@@ -1,7 +1,12 @@
 import { useRecoilValue } from 'recoil';
 import { useCallback, useRef, useEffect } from 'react';
 import { useGetModelsQuery } from 'librechat-data-provider/react-query';
-import { getEndpointField, LocalStorageKeys, isAssistantsEndpoint } from 'librechat-data-provider';
+import {
+  getEndpointField,
+  LocalStorageKeys,
+  isAssistantsEndpoint,
+  getDefaultParamsEndpoint,
+} from 'librechat-data-provider';
 import type {
   TEndpointsConfig,
   EModelEndpoint,
@@ -117,11 +122,13 @@ const useGenerateConvo = ({
       }
 
       const models = modelsConfig?.[defaultEndpoint ?? ''] ?? [];
+      const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, defaultEndpoint);
       conversation = buildDefaultConvo({
         conversation,
         lastConversationSetup: preset as TConversation,
         endpoint: defaultEndpoint ?? ('' as EModelEndpoint),
         models,
+        defaultParamsEndpoint,
       });
 
       if (preset?.title != null && preset.title !== '') {
diff --git a/client/src/hooks/Conversations/useNavigateToConvo.tsx b/client/src/hooks/Conversations/useNavigateToConvo.tsx
index 114b70c6ef..b9d188eaf0 100644
--- a/client/src/hooks/Conversations/useNavigateToConvo.tsx
+++ b/client/src/hooks/Conversations/useNavigateToConvo.tsx
@@ -2,7 +2,13 @@ import { useCallback } from 'react';
 import { useSetRecoilState } from 'recoil';
 import { useNavigate } from 'react-router-dom';
 import { useQueryClient } from '@tanstack/react-query';
-import { QueryKeys, Constants, dataService, getEndpointField } from 'librechat-data-provider';
+import {
+  QueryKeys,
+  Constants,
+  dataService,
+  getEndpointField,
+  getDefaultParamsEndpoint,
+} from 'librechat-data-provider';
 import type {
   TEndpointsConfig,
   TStartupConfig,
@@ -106,11 +112,13 @@ const useNavigateToConvo = (index = 0) => {
 
       const models = modelsConfig?.[defaultEndpoint ?? ''] ?? [];
 
+      const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, defaultEndpoint);
       convo = buildDefaultConvo({
         models,
         conversation,
         endpoint: defaultEndpoint,
         lastConversationSetup: conversation,
+        defaultParamsEndpoint,
       });
     }
     clearAllConversations(true);
diff --git a/client/src/hooks/useNewConvo.ts b/client/src/hooks/useNewConvo.ts
index c468ab30a2..7fa499f40d 100644
--- a/client/src/hooks/useNewConvo.ts
+++ b/client/src/hooks/useNewConvo.ts
@@ -14,6 +14,7 @@ import {
   LocalStorageKeys,
   isEphemeralAgentId,
   isAssistantsEndpoint,
+  getDefaultParamsEndpoint,
 } from 'librechat-data-provider';
 import type {
   TPreset,
@@ -191,11 +192,13 @@ const useNewConvo = (index = 0) => {
           }
 
           const models = modelsConfig?.[defaultEndpoint] ?? [];
+          const defaultParamsEndpoint = getDefaultParamsEndpoint(endpointsConfig, defaultEndpoint);
           conversation = buildDefaultConvo({
             conversation,
             lastConversationSetup: activePreset as TConversation,
             endpoint: defaultEndpoint,
             models,
+            defaultParamsEndpoint,
           });
         }
 
diff --git a/client/src/utils/__tests__/buildDefaultConvo.test.ts b/client/src/utils/__tests__/buildDefaultConvo.test.ts
new file mode 100644
index 0000000000..00a4d6313b
--- /dev/null
+++ b/client/src/utils/__tests__/buildDefaultConvo.test.ts
@@ -0,0 +1,202 @@
+import { EModelEndpoint } from 'librechat-data-provider';
+import type { TConversation } from 'librechat-data-provider';
+import buildDefaultConvo from '../buildDefaultConvo';
+
+jest.mock('../localStorage', () => ({
+  getLocalStorageItems: jest.fn(() => ({
+    lastSelectedModel: {},
+    lastSelectedTools: [],
+    lastConversationSetup: {},
+  })),
+}));
+
+const baseConversation: TConversation = {
+  conversationId: 'test-convo-id',
+  title: 'Test Conversation',
+  createdAt: '2024-01-01T00:00:00Z',
+  updatedAt: '2024-01-01T00:00:00Z',
+  endpoint: null,
+};
+
+describe('buildDefaultConvo - defaultParamsEndpoint', () => {
+  describe('custom endpoint with defaultParamsEndpoint: anthropic', () => {
+    const models = ['anthropic/claude-opus-4.5', 'anthropic/claude-sonnet-4'];
+
+    it('should preserve maxOutputTokens from model spec preset', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        topP: 0.9,
+        maxContextTokens: 50000,
+      };
+
+      const result = buildDefaultConvo({
+        models,
+        conversation: baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        lastConversationSetup: preset,
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      });
+
+      expect(result.maxOutputTokens).toBe(8192);
+      expect(result.topP).toBe(0.9);
+      expect(result.temperature).toBe(0.7);
+      expect(result.maxContextTokens).toBe(50000);
+      expect(result.model).toBe('anthropic/claude-opus-4.5');
+    });
+
+    it('should strip maxOutputTokens without defaultParamsEndpoint', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+      };
+
+      const result = buildDefaultConvo({
+        models,
+        conversation: baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        lastConversationSetup: preset,
+      });
+
+      expect(result.maxOutputTokens).toBeUndefined();
+      expect(result.temperature).toBe(0.7);
+    });
+
+    it('should strip OpenAI-specific fields when using anthropic params', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'anthropic/claude-opus-4.5',
+        max_tokens: 4096,
+        top_p: 0.9,
+        presence_penalty: 0.5,
+        frequency_penalty: 0.3,
+      };
+
+      const result = buildDefaultConvo({
+        models,
+        conversation: baseConversation,
+        endpoint: 'AnthropicClaude' as EModelEndpoint,
+        lastConversationSetup: preset,
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      });
+
+      expect(result.max_tokens).toBeUndefined();
+      expect(result.top_p).toBeUndefined();
+      expect(result.presence_penalty).toBeUndefined();
+      expect(result.frequency_penalty).toBeUndefined();
+    });
+  });
+
+  describe('custom endpoint without defaultParamsEndpoint (OpenAI default)', () => {
+    const models = ['gpt-4o', 'gpt-4.1'];
+
+    it('should preserve OpenAI fields and strip anthropic fields', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'MyOpenRouterEndpoint' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'gpt-4o',
+        temperature: 0.7,
+        max_tokens: 4096,
+        top_p: 0.9,
+        maxOutputTokens: 8192,
+      };
+
+      const result = buildDefaultConvo({
+        models,
+        conversation: baseConversation,
+        endpoint: 'MyOpenRouterEndpoint' as EModelEndpoint,
+        lastConversationSetup: preset,
+      });
+
+      expect(result.max_tokens).toBe(4096);
+      expect(result.top_p).toBe(0.9);
+      expect(result.temperature).toBe(0.7);
+      expect(result.maxOutputTokens).toBeUndefined();
+    });
+  });
+
+  describe('custom endpoint with defaultParamsEndpoint: google', () => {
+    const models = ['gemini-pro', 'gemini-1.5-pro'];
+
+    it('should preserve Google-specific fields', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'MyGoogleEndpoint' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'gemini-pro',
+        temperature: 0.7,
+        maxOutputTokens: 8192,
+        topP: 0.9,
+        topK: 40,
+      };
+
+      const result = buildDefaultConvo({
+        models,
+        conversation: baseConversation,
+        endpoint: 'MyGoogleEndpoint' as EModelEndpoint,
+        lastConversationSetup: preset,
+        defaultParamsEndpoint: EModelEndpoint.google,
+      });
+
+      expect(result.maxOutputTokens).toBe(8192);
+      expect(result.topP).toBe(0.9);
+      expect(result.topK).toBe(40);
+    });
+  });
+
+  describe('cross-endpoint field isolation', () => {
+    it('should not carry bedrock region to a custom endpoint', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'MyChatEndpoint' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'gpt-4o',
+        temperature: 0.7,
+        region: 'us-east-1',
+      };
+
+      const result = buildDefaultConvo({
+        models: ['gpt-4o'],
+        conversation: baseConversation,
+        endpoint: 'MyChatEndpoint' as EModelEndpoint,
+        lastConversationSetup: preset,
+      });
+
+      expect(result.region).toBeUndefined();
+      expect(result.temperature).toBe(0.7);
+    });
+
+    it('should not carry bedrock region even with anthropic defaultParamsEndpoint', () => {
+      const preset: TConversation = {
+        ...baseConversation,
+        endpoint: 'MyChatEndpoint' as EModelEndpoint,
+        endpointType: EModelEndpoint.custom,
+        model: 'claude-3-opus',
+        region: 'us-east-1',
+        maxOutputTokens: 8192,
+      };
+
+      const result = buildDefaultConvo({
+        models: ['claude-3-opus'],
+        conversation: baseConversation,
+        endpoint: 'MyChatEndpoint' as EModelEndpoint,
+        lastConversationSetup: preset,
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      });
+
+      expect(result.region).toBeUndefined();
+      expect(result.maxOutputTokens).toBe(8192);
+    });
+  });
+});
diff --git a/client/src/utils/__tests__/cleanupPreset.integration.test.ts b/client/src/utils/__tests__/cleanupPreset.integration.test.ts
new file mode 100644
index 0000000000..1e1219bc7a
--- /dev/null
+++ b/client/src/utils/__tests__/cleanupPreset.integration.test.ts
@@ -0,0 +1,119 @@
+import { EModelEndpoint } from 'librechat-data-provider';
+import cleanupPreset from '../cleanupPreset';
+
+/**
+ * Integration tests for cleanupPreset — NO mocks.
+ * Uses the real parseConvo to verify actual schema behavior
+ * with defaultParamsEndpoint for custom endpoints.
+ */
+describe('cleanupPreset - real parsing with defaultParamsEndpoint', () => {
+  it('should preserve maxOutputTokens when defaultParamsEndpoint is anthropic', () => {
+    const preset = {
+      presetId: 'test-id',
+      title: 'Claude Opus',
+      endpoint: 'AnthropicClaude',
+      endpointType: EModelEndpoint.custom,
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      topP: 0.9,
+      maxContextTokens: 50000,
+    };
+
+    const result = cleanupPreset({
+      preset,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result.maxOutputTokens).toBe(8192);
+    expect(result.topP).toBe(0.9);
+    expect(result.temperature).toBe(0.7);
+    expect(result.maxContextTokens).toBe(50000);
+    expect(result.model).toBe('anthropic/claude-opus-4.5');
+  });
+
+  it('should strip maxOutputTokens without defaultParamsEndpoint (OpenAI schema)', () => {
+    const preset = {
+      presetId: 'test-id',
+      title: 'GPT Custom',
+      endpoint: 'MyOpenRouter',
+      endpointType: EModelEndpoint.custom,
+      model: 'gpt-4o',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      max_tokens: 4096,
+    };
+
+    const result = cleanupPreset({ preset });
+
+    expect(result.maxOutputTokens).toBeUndefined();
+    expect(result.max_tokens).toBe(4096);
+    expect(result.temperature).toBe(0.7);
+  });
+
+  it('should strip OpenAI-specific fields when using anthropic params', () => {
+    const preset = {
+      presetId: 'test-id',
+      title: 'Claude Custom',
+      endpoint: 'AnthropicClaude',
+      endpointType: EModelEndpoint.custom,
+      model: 'anthropic/claude-3-opus',
+      max_tokens: 4096,
+      top_p: 0.9,
+      presence_penalty: 0.5,
+      frequency_penalty: 0.3,
+      temperature: 0.7,
+    };
+
+    const result = cleanupPreset({
+      preset,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result.max_tokens).toBeUndefined();
+    expect(result.top_p).toBeUndefined();
+    expect(result.presence_penalty).toBeUndefined();
+    expect(result.frequency_penalty).toBeUndefined();
+    expect(result.temperature).toBe(0.7);
+  });
+
+  it('should not carry bedrock region to custom endpoint', () => {
+    const preset = {
+      presetId: 'test-id',
+      title: 'Custom',
+      endpoint: 'MyEndpoint',
+      endpointType: EModelEndpoint.custom,
+      model: 'gpt-4o',
+      temperature: 0.7,
+      region: 'us-east-1',
+    };
+
+    const result = cleanupPreset({ preset });
+
+    expect(result.region).toBeUndefined();
+    expect(result.temperature).toBe(0.7);
+  });
+
+  it('should preserve Google-specific fields when defaultParamsEndpoint is google', () => {
+    const preset = {
+      presetId: 'test-id',
+      title: 'Gemini Custom',
+      endpoint: 'MyGoogleEndpoint',
+      endpointType: EModelEndpoint.custom,
+      model: 'gemini-pro',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      topP: 0.9,
+      topK: 40,
+    };
+
+    const result = cleanupPreset({
+      preset,
+      defaultParamsEndpoint: EModelEndpoint.google,
+    });
+
+    expect(result.maxOutputTokens).toBe(8192);
+    expect(result.topP).toBe(0.9);
+    expect(result.topK).toBe(40);
+  });
+});
diff --git a/client/src/utils/__tests__/cleanupPreset.test.ts b/client/src/utils/__tests__/cleanupPreset.test.ts
index a03477de15..766bb872ac 100644
--- a/client/src/utils/__tests__/cleanupPreset.test.ts
+++ b/client/src/utils/__tests__/cleanupPreset.test.ts
@@ -1,12 +1,9 @@
-import { EModelEndpoint } from 'librechat-data-provider';
+import { EModelEndpoint, parseConvo } from 'librechat-data-provider';
 import cleanupPreset from '../cleanupPreset';
-import type { TPreset } from 'librechat-data-provider';
-
 // Mock parseConvo since we're focusing on testing the chatGptLabel migration logic
 jest.mock('librechat-data-provider', () => ({
   ...jest.requireActual('librechat-data-provider'),
   parseConvo: jest.fn((input) => {
-    // Return a simplified mock that passes through most properties
     const { conversation } = input;
     return {
       ...conversation,
@@ -221,4 +218,41 @@ describe('cleanupPreset', () => {
       expect(result.presetId).toBeNull();
     });
   });
+
+  describe('defaultParamsEndpoint threading', () => {
+    it('should pass defaultParamsEndpoint to parseConvo', () => {
+      const preset = {
+        ...basePreset,
+        endpoint: 'MyCustomEndpoint',
+        endpointType: EModelEndpoint.custom,
+      };
+
+      cleanupPreset({
+        preset,
+        defaultParamsEndpoint: EModelEndpoint.anthropic,
+      });
+
+      expect(parseConvo).toHaveBeenCalledWith(
+        expect.objectContaining({
+          defaultParamsEndpoint: EModelEndpoint.anthropic,
+        }),
+      );
+    });
+
+    it('should pass undefined defaultParamsEndpoint when not provided', () => {
+      const preset = {
+        ...basePreset,
+        endpoint: 'MyCustomEndpoint',
+        endpointType: EModelEndpoint.custom,
+      };
+
+      cleanupPreset({ preset });
+
+      expect(parseConvo).toHaveBeenCalledWith(
+        expect.objectContaining({
+          defaultParamsEndpoint: undefined,
+        }),
+      );
+    });
+  });
 });
diff --git a/client/src/utils/buildDefaultConvo.ts b/client/src/utils/buildDefaultConvo.ts
index 025bec24eb..c2d2871912 100644
--- a/client/src/utils/buildDefaultConvo.ts
+++ b/client/src/utils/buildDefaultConvo.ts
@@ -14,11 +14,13 @@ const buildDefaultConvo = ({
   conversation,
   endpoint = null,
   lastConversationSetup,
+  defaultParamsEndpoint,
 }: {
   models: string[];
   conversation: TConversation;
   endpoint?: EModelEndpoint | null;
   lastConversationSetup: TConversation | null;
+  defaultParamsEndpoint?: string | null;
 }): TConversation => {
   const { lastSelectedModel, lastSelectedTools } = getLocalStorageItems();
   const endpointType = lastConversationSetup?.endpointType ?? conversation.endpointType;
@@ -49,6 +51,7 @@ const buildDefaultConvo = ({
     possibleValues: {
       models: possibleModels,
     },
+    defaultParamsEndpoint,
   });
 
   const defaultConvo = {
diff --git a/client/src/utils/cleanupPreset.ts b/client/src/utils/cleanupPreset.ts
index c158d935fa..ad44726064 100644
--- a/client/src/utils/cleanupPreset.ts
+++ b/client/src/utils/cleanupPreset.ts
@@ -4,9 +4,10 @@ import type { TPreset } from 'librechat-data-provider';
 type UIPreset = Partial<TPreset> & { presetOverride?: Partial<TPreset> };
 type TCleanupPreset = {
   preset?: UIPreset;
+  defaultParamsEndpoint?: string | null;
 };
 
-const cleanupPreset = ({ preset: _preset }: TCleanupPreset): TPreset => {
+const cleanupPreset = ({ preset: _preset, defaultParamsEndpoint }: TCleanupPreset): TPreset => {
   const { endpoint, endpointType } = _preset ?? ({} as UIPreset);
   if (endpoint == null || endpoint === '') {
     console.error(`Unknown endpoint ${endpoint}`, _preset);
@@ -35,8 +36,13 @@ const cleanupPreset = ({ preset: _preset }: TCleanupPreset): TPreset => {
     delete preset.chatGptLabel;
   }
 
-  /* @ts-ignore: endpoint can be a custom defined name */
-  const parsedPreset = parseConvo({ endpoint, endpointType, conversation: preset });
+  const parsedPreset = parseConvo({
+    /* @ts-ignore: endpoint can be a custom defined name */
+    endpoint,
+    endpointType,
+    conversation: preset,
+    defaultParamsEndpoint,
+  });
 
   return {
     presetId: _preset?.presetId ?? null,
diff --git a/packages/api/src/agents/__tests__/initialize.test.ts b/packages/api/src/agents/__tests__/initialize.test.ts
new file mode 100644
index 0000000000..01310a09c4
--- /dev/null
+++ b/packages/api/src/agents/__tests__/initialize.test.ts
@@ -0,0 +1,284 @@
+import { Providers } from '@librechat/agents';
+import { EModelEndpoint } from 'librechat-data-provider';
+import type { Agent } from 'librechat-data-provider';
+import type { ServerRequest, InitializeResultBase } from '~/types';
+import type { InitializeAgentDbMethods } from '../initialize';
+
+// Mock logger
+jest.mock('winston', () => ({
+  createLogger: jest.fn(() => ({
+    debug: jest.fn(),
+    warn: jest.fn(),
+    error: jest.fn(),
+  })),
+  format: {
+    combine: jest.fn(),
+    colorize: jest.fn(),
+    simple: jest.fn(),
+  },
+  transports: {
+    Console: jest.fn(),
+  },
+}));
+
+const mockExtractLibreChatParams = jest.fn();
+const mockGetModelMaxTokens = jest.fn();
+const mockOptionalChainWithEmptyCheck = jest.fn();
+const mockGetThreadData = jest.fn();
+
+jest.mock('~/utils', () => ({
+  extractLibreChatParams: (...args: unknown[]) => mockExtractLibreChatParams(...args),
+  getModelMaxTokens: (...args: unknown[]) => mockGetModelMaxTokens(...args),
+  optionalChainWithEmptyCheck: (...args: unknown[]) => mockOptionalChainWithEmptyCheck(...args),
+  getThreadData: (...args: unknown[]) => mockGetThreadData(...args),
+}));
+
+const mockGetProviderConfig = jest.fn();
+jest.mock('~/endpoints', () => ({
+  getProviderConfig: (...args: unknown[]) => mockGetProviderConfig(...args),
+}));
+
+jest.mock('~/files', () => ({
+  filterFilesByEndpointConfig: jest.fn(() => []),
+}));
+
+jest.mock('~/prompts', () => ({
+  generateArtifactsPrompt: jest.fn(() => null),
+}));
+
+jest.mock('../resources', () => ({
+  primeResources: jest.fn().mockResolvedValue({
+    attachments: [],
+    tool_resources: undefined,
+  }),
+}));
+
+import { initializeAgent } from '../initialize';
+
+/**
+ * Creates minimal mock objects for initializeAgent tests.
+ */
+function createMocks(overrides?: {
+  maxContextTokens?: number;
+  modelDefault?: number;
+  maxOutputTokens?: number;
+}) {
+  const { maxContextTokens, modelDefault = 200000, maxOutputTokens = 4096 } = overrides ?? {};
+
+  const agent = {
+    id: 'agent-1',
+    model: 'test-model',
+    provider: Providers.OPENAI,
+    tools: [],
+    model_parameters: { model: 'test-model' },
+  } as unknown as Agent;
+
+  const req = {
+    user: { id: 'user-1' },
+    config: {},
+  } as unknown as ServerRequest;
+
+  const res = {} as unknown as import('express').Response;
+
+  const mockGetOptions = jest.fn().mockResolvedValue({
+    llmConfig: {
+      model: 'test-model',
+      maxTokens: maxOutputTokens,
+    },
+    endpointTokenConfig: undefined,
+  } satisfies InitializeResultBase);
+
+  mockGetProviderConfig.mockReturnValue({
+    getOptions: mockGetOptions,
+    overrideProvider: Providers.OPENAI,
+  });
+
+  // extractLibreChatParams returns maxContextTokens when provided in model_parameters
+  mockExtractLibreChatParams.mockReturnValue({
+    resendFiles: false,
+    maxContextTokens,
+    modelOptions: { model: 'test-model' },
+  });
+
+  // getModelMaxTokens returns the model's default context window
+  mockGetModelMaxTokens.mockReturnValue(modelDefault);
+
+  // Implement real optionalChainWithEmptyCheck behavior
+  mockOptionalChainWithEmptyCheck.mockImplementation(
+    (...values: (string | number | undefined)[]) => {
+      for (const v of values) {
+        if (v !== undefined && v !== null && v !== '') {
+          return v;
+        }
+      }
+      return values[values.length - 1];
+    },
+  );
+
+  const loadTools = jest.fn().mockResolvedValue({
+    tools: [],
+    toolContextMap: {},
+    userMCPAuthMap: undefined,
+    toolRegistry: undefined,
+    toolDefinitions: [],
+    hasDeferredTools: false,
+  });
+
+  const db: InitializeAgentDbMethods = {
+    getFiles: jest.fn().mockResolvedValue([]),
+    getConvoFiles: jest.fn().mockResolvedValue([]),
+    updateFilesUsage: jest.fn().mockResolvedValue([]),
+    getUserKey: jest.fn().mockResolvedValue('user-1'),
+    getUserKeyValues: jest.fn().mockResolvedValue([]),
+    getToolFilesByIds: jest.fn().mockResolvedValue([]),
+  };
+
+  return { agent, req, res, loadTools, db };
+}
+
+describe('initializeAgent — maxContextTokens', () => {
+  beforeEach(() => {
+    jest.clearAllMocks();
+  });
+
+  it('uses user-configured maxContextTokens when provided via model_parameters', async () => {
+    const userValue = 50000;
+    const { agent, req, res, loadTools, db } = createMocks({
+      maxContextTokens: userValue,
+      modelDefault: 200000,
+      maxOutputTokens: 4096,
+    });
+
+    const result = await initializeAgent(
+      {
+        req,
+        res,
+        agent,
+        loadTools,
+        endpointOption: {
+          endpoint: EModelEndpoint.agents,
+          model_parameters: { maxContextTokens: userValue },
+        },
+        allowedProviders: new Set([Providers.OPENAI]),
+        isInitialAgent: true,
+      },
+      db,
+    );
+
+    expect(result.maxContextTokens).toBe(userValue);
+  });
+
+  it('falls back to formula when maxContextTokens is NOT provided', async () => {
+    const modelDefault = 200000;
+    const maxOutputTokens = 4096;
+    const { agent, req, res, loadTools, db } = createMocks({
+      maxContextTokens: undefined,
+      modelDefault,
+      maxOutputTokens,
+    });
+
+    const result = await initializeAgent(
+      {
+        req,
+        res,
+        agent,
+        loadTools,
+        endpointOption: { endpoint: EModelEndpoint.agents },
+        allowedProviders: new Set([Providers.OPENAI]),
+        isInitialAgent: true,
+      },
+      db,
+    );
+
+    const expected = Math.round((modelDefault - maxOutputTokens) * 0.9);
+    expect(result.maxContextTokens).toBe(expected);
+  });
+
+  it('falls back to formula when maxContextTokens is 0', async () => {
+    const maxOutputTokens = 4096;
+    const { agent, req, res, loadTools, db } = createMocks({
+      maxContextTokens: 0,
+      modelDefault: 200000,
+      maxOutputTokens,
+    });
+
+    const result = await initializeAgent(
+      {
+        req,
+        res,
+        agent,
+        loadTools,
+        endpointOption: {
+          endpoint: EModelEndpoint.agents,
+          model_parameters: { maxContextTokens: 0 },
+        },
+        allowedProviders: new Set([Providers.OPENAI]),
+        isInitialAgent: true,
+      },
+      db,
+    );
+
+    // 0 is not used as-is; the formula kicks in.
+    // optionalChainWithEmptyCheck(0, 200000, 18000) returns 0 (not null/undefined),
+    // then Number(0) || 18000 = 18000 (the fallback default).
+    expect(result.maxContextTokens).not.toBe(0);
+    const expected = Math.round((18000 - maxOutputTokens) * 0.9);
+    expect(result.maxContextTokens).toBe(expected);
+  });
+
+  it('falls back to formula when maxContextTokens is negative', async () => {
+    const maxOutputTokens = 4096;
+    const { agent, req, res, loadTools, db } = createMocks({
+      maxContextTokens: -1,
+      modelDefault: 200000,
+      maxOutputTokens,
+    });
+
+    const result = await initializeAgent(
+      {
+        req,
+        res,
+        agent,
+        loadTools,
+        endpointOption: {
+          endpoint: EModelEndpoint.agents,
+          model_parameters: { maxContextTokens: -1 },
+        },
+        allowedProviders: new Set([Providers.OPENAI]),
+        isInitialAgent: true,
+      },
+      db,
+    );
+
+    // -1 is not used as-is; the formula kicks in
+    expect(result.maxContextTokens).not.toBe(-1);
+  });
+
+  it('preserves small user-configured value (e.g. 1000 from modelSpec)', async () => {
+    const userValue = 1000;
+    const { agent, req, res, loadTools, db } = createMocks({
+      maxContextTokens: userValue,
+      modelDefault: 128000,
+      maxOutputTokens: 4096,
+    });
+
+    const result = await initializeAgent(
+      {
+        req,
+        res,
+        agent,
+        loadTools,
+        endpointOption: {
+          endpoint: EModelEndpoint.agents,
+          model_parameters: { maxContextTokens: userValue },
+        },
+        allowedProviders: new Set([Providers.OPENAI]),
+        isInitialAgent: true,
+      },
+      db,
+    );
+
+    // Should NOT be overridden to Math.round((128000 - 4096) * 0.9) = 111,514
+    expect(result.maxContextTokens).toBe(userValue);
+  });
+});
diff --git a/packages/api/src/agents/initialize.ts b/packages/api/src/agents/initialize.ts
index 008aa4c0ba..af604beb81 100644
--- a/packages/api/src/agents/initialize.ts
+++ b/packages/api/src/agents/initialize.ts
@@ -413,7 +413,10 @@ export async function initializeAgent(
     toolContextMap: toolContextMap ?? {},
     useLegacyContent: !!options.useLegacyContent,
     tools: (tools ?? []) as GenericTool[] & string[],
-    maxContextTokens: Math.round((agentMaxContextNum - maxOutputTokensNum) * 0.9),
+    maxContextTokens:
+      maxContextTokens != null && maxContextTokens > 0
+        ? maxContextTokens
+        : Math.round((agentMaxContextNum - maxOutputTokensNum) * 0.9),
   };
 
   return initializedAgent;
diff --git a/packages/data-provider/specs/parsers.spec.ts b/packages/data-provider/specs/parsers.spec.ts
index 012d33177b..90359d6e6f 100644
--- a/packages/data-provider/specs/parsers.spec.ts
+++ b/packages/data-provider/specs/parsers.spec.ts
@@ -1,4 +1,4 @@
-import { replaceSpecialVars, parseCompactConvo, parseTextParts } from '../src/parsers';
+import { replaceSpecialVars, parseConvo, parseCompactConvo, parseTextParts } from '../src/parsers';
 import { specialVariables } from '../src/config';
 import { EModelEndpoint } from '../src/schemas';
 import { ContentTypes } from '../src/types/runs';
@@ -262,6 +262,257 @@ describe('parseCompactConvo', () => {
   });
 });
 
+describe('parseConvo - defaultParamsEndpoint', () => {
+  test('should strip maxOutputTokens for custom endpoint without defaultParamsEndpoint', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      maxContextTokens: 50000,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.temperature).toBe(0.7);
+    expect(result?.maxContextTokens).toBe(50000);
+    expect(result?.maxOutputTokens).toBeUndefined();
+  });
+
+  test('should preserve maxOutputTokens when defaultParamsEndpoint is anthropic', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      topP: 0.9,
+      topK: 40,
+      maxContextTokens: 50000,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.model).toBe('anthropic/claude-opus-4.5');
+    expect(result?.temperature).toBe(0.7);
+    expect(result?.maxOutputTokens).toBe(8192);
+    expect(result?.topP).toBe(0.9);
+    expect(result?.topK).toBe(40);
+    expect(result?.maxContextTokens).toBe(50000);
+  });
+
+  test('should strip OpenAI-specific fields when defaultParamsEndpoint is anthropic', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      max_tokens: 4096,
+      top_p: 0.9,
+      presence_penalty: 0.5,
+      frequency_penalty: 0.3,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.temperature).toBe(0.7);
+    expect(result?.max_tokens).toBeUndefined();
+    expect(result?.top_p).toBeUndefined();
+    expect(result?.presence_penalty).toBeUndefined();
+    expect(result?.frequency_penalty).toBeUndefined();
+  });
+
+  test('should preserve max_tokens when defaultParamsEndpoint is not set (OpenAI default)', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gpt-4o',
+      temperature: 0.7,
+      max_tokens: 4096,
+      top_p: 0.9,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.max_tokens).toBe(4096);
+    expect(result?.top_p).toBe(0.9);
+  });
+
+  test('should preserve Google-specific fields when defaultParamsEndpoint is google', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gemini-pro',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      topP: 0.9,
+      topK: 40,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.google,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.maxOutputTokens).toBe(8192);
+    expect(result?.topP).toBe(0.9);
+    expect(result?.topK).toBe(40);
+  });
+
+  test('should not strip fields from non-custom endpoints that already have a schema', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gpt-4o',
+      temperature: 0.7,
+      max_tokens: 4096,
+      top_p: 0.9,
+    };
+
+    const result = parseConvo({
+      endpoint: EModelEndpoint.openAI,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.max_tokens).toBe(4096);
+    expect(result?.top_p).toBe(0.9);
+  });
+
+  test('should not carry bedrock region to custom endpoint without defaultParamsEndpoint', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gpt-4o',
+      temperature: 0.7,
+      region: 'us-east-1',
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.temperature).toBe(0.7);
+    expect(result?.region).toBeUndefined();
+  });
+
+  test('should fall back to endpointType schema when defaultParamsEndpoint is invalid', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gpt-4o',
+      temperature: 0.7,
+      max_tokens: 4096,
+      maxOutputTokens: 8192,
+    };
+
+    const result = parseConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: 'nonexistent_endpoint',
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.max_tokens).toBe(4096);
+    expect(result?.maxOutputTokens).toBeUndefined();
+  });
+});
+
+describe('parseCompactConvo - defaultParamsEndpoint', () => {
+  test('should strip maxOutputTokens for custom endpoint without defaultParamsEndpoint', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+    };
+
+    const result = parseCompactConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.temperature).toBe(0.7);
+    expect(result?.maxOutputTokens).toBeUndefined();
+  });
+
+  test('should preserve maxOutputTokens when defaultParamsEndpoint is anthropic', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      temperature: 0.7,
+      maxOutputTokens: 8192,
+      topP: 0.9,
+      maxContextTokens: 50000,
+    };
+
+    const result = parseCompactConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.maxOutputTokens).toBe(8192);
+    expect(result?.topP).toBe(0.9);
+    expect(result?.maxContextTokens).toBe(50000);
+  });
+
+  test('should strip iconURL even when defaultParamsEndpoint is set', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'anthropic/claude-opus-4.5',
+      iconURL: 'https://malicious.com/track.png',
+      maxOutputTokens: 8192,
+    };
+
+    const result = parseCompactConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: EModelEndpoint.anthropic,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.['iconURL']).toBeUndefined();
+    expect(result?.maxOutputTokens).toBe(8192);
+  });
+
+  test('should fall back to endpointType when defaultParamsEndpoint is null', () => {
+    const conversation: Partial<TConversation> = {
+      model: 'gpt-4o',
+      max_tokens: 4096,
+      maxOutputTokens: 8192,
+    };
+
+    const result = parseCompactConvo({
+      endpoint: 'MyCustomEndpoint' as EModelEndpoint,
+      endpointType: EModelEndpoint.custom,
+      conversation,
+      defaultParamsEndpoint: null,
+    });
+
+    expect(result).not.toBeNull();
+    expect(result?.max_tokens).toBe(4096);
+    expect(result?.maxOutputTokens).toBeUndefined();
+  });
+});
+
 describe('parseTextParts', () => {
   test('should concatenate text parts', () => {
     const parts: TMessageContentParts[] = [
diff --git a/packages/data-provider/src/config.ts b/packages/data-provider/src/config.ts
index a2b47351b1..f6567e8da9 100644
--- a/packages/data-provider/src/config.ts
+++ b/packages/data-provider/src/config.ts
@@ -1908,3 +1908,14 @@ export function getEndpointField<
   }
   return config[property];
 }
+
+/** Resolves the `defaultParamsEndpoint` for a given endpoint from its custom params config */
+export function getDefaultParamsEndpoint(
+  endpointsConfig: TEndpointsConfig | undefined | null,
+  endpoint: string | null | undefined,
+): string | undefined {
+  if (!endpointsConfig || !endpoint) {
+    return undefined;
+  }
+  return endpointsConfig[endpoint]?.customParams?.defaultParamsEndpoint;
+}
diff --git a/packages/data-provider/src/parsers.ts b/packages/data-provider/src/parsers.ts
index ff47375769..ab6ff927f7 100644
--- a/packages/data-provider/src/parsers.ts
+++ b/packages/data-provider/src/parsers.ts
@@ -144,26 +144,25 @@ export const parseConvo = ({
   endpointType,
   conversation,
   possibleValues,
+  defaultParamsEndpoint,
 }: {
   endpoint: EndpointSchemaKey;
   endpointType?: EndpointSchemaKey | null;
   conversation: Partial<s.TConversation | s.TPreset> | null;
   possibleValues?: TPossibleValues;
-  // TODO: POC for default schema
-  // defaultSchema?: Partial<EndpointSchema>,
+  defaultParamsEndpoint?: string | null;
 }) => {
   let schema = endpointSchemas[endpoint] as EndpointSchema | undefined;
 
   if (!schema && !endpointType) {
     throw new Error(`Unknown endpoint: ${endpoint}`);
-  } else if (!schema && endpointType) {
-    schema = endpointSchemas[endpointType];
+  } else if (!schema) {
+    const overrideSchema = defaultParamsEndpoint
+      ? endpointSchemas[defaultParamsEndpoint as EndpointSchemaKey]
+      : undefined;
+    schema = overrideSchema ?? (endpointType ? endpointSchemas[endpointType] : undefined);
   }
 
-  // if (defaultSchema && schemaCreators[endpoint]) {
-  //   schema = schemaCreators[endpoint](defaultSchema);
-  // }
-
   const convo = schema?.parse(conversation) as s.TConversation | undefined;
   const { models } = possibleValues ?? {};
 
@@ -310,13 +309,13 @@ export const parseCompactConvo = ({
   endpointType,
   conversation,
   possibleValues,
+  defaultParamsEndpoint,
 }: {
   endpoint?: EndpointSchemaKey;
   endpointType?: EndpointSchemaKey | null;
   conversation: Partial<s.TConversation | s.TPreset>;
   possibleValues?: TPossibleValues;
-  // TODO: POC for default schema
-  // defaultSchema?: Partial<EndpointSchema>,
+  defaultParamsEndpoint?: string | null;
 }): Omit<s.TConversation, 'iconURL'> | null => {
   if (!endpoint) {
     throw new Error(`undefined endpoint: ${endpoint}`);
@@ -326,8 +325,11 @@ export const parseCompactConvo = ({
 
   if (!schema && !endpointType) {
     throw new Error(`Unknown endpoint: ${endpoint}`);
-  } else if (!schema && endpointType) {
-    schema = compactEndpointSchemas[endpointType];
+  } else if (!schema) {
+    const overrideSchema = defaultParamsEndpoint
+      ? compactEndpointSchemas[defaultParamsEndpoint as EndpointSchemaKey]
+      : undefined;
+    schema = overrideSchema ?? (endpointType ? compactEndpointSchemas[endpointType] : undefined);
   }
 
   if (!schema) {