diff --git a/packages/api/src/endpoints/custom/config.spec.ts b/packages/api/src/endpoints/custom/config.spec.ts
index b0154b3b41b..8fad8b31ed2 100644
--- a/packages/api/src/endpoints/custom/config.spec.ts
+++ b/packages/api/src/endpoints/custom/config.spec.ts
@@ -1,4 +1,4 @@
-import { AuthType, EModelEndpoint } from 'librechat-data-provider';
+import { AuthType, EModelEndpoint, ReasoningParameterFormat } from 'librechat-data-provider';
 import type { TCustomEndpoints } from 'librechat-data-provider';
 import { loadCustomEndpointsConfig } from './config';
 
@@ -43,6 +43,103 @@ describe('loadCustomEndpointsConfig – native provider param set', () => {
   });
 });
 
+describe('loadCustomEndpointsConfig: host-implied reasoning support', () => {
+  const load = (endpoint: Record<string, unknown>) =>
+    loadCustomEndpointsConfig([
+      { ...baseEndpoint, name: 'Gateway', ...endpoint },
+    ] as unknown as TCustomEndpoints)?.Gateway?.customParams;
+
+  it('declares effort-style reasoning for OpenRouter', () => {
+    expect(load({ baseURL: 'https://openrouter.ai/api/v1' })?.reasoningFormat).toBe(
+      ReasoningParameterFormat.reasoningEffort,
+    );
+  });
+
+  it.each(['https://api.openai.com/v1', 'https://api.x.ai/v1'])(
+    'leaves %s to explicit config: its capability depends on the model',
+    (baseURL) => {
+      expect(load({ baseURL })).toBeUndefined();
+    },
+  );
+
+  it.each([['reasoning_effort'], ['reasoning']])(
+    'does not advertise reasoning that dropParams %j removes',
+    (...dropParams) => {
+      expect(load({ baseURL: 'https://openrouter.ai/api/v1', dropParams })).toBeUndefined();
+    },
+  );
+
+  it('advertises reasoning for the OpenRouter params endpoint the config loader injects', () => {
+    expect(
+      load({
+        baseURL: 'https://openrouter.ai/api/v1',
+        customParams: {
+          defaultParamsEndpoint: 'openrouter',
+          paramDefinitions: [{ key: 'promptCache', default: true }],
+        },
+      }),
+    ).toEqual({
+      defaultParamsEndpoint: 'openrouter',
+      paramDefinitions: [{ key: 'promptCache', default: true }],
+      reasoningFormat: ReasoningParameterFormat.reasoningEffort,
+    });
+  });
+
+  it('still advertises reasoning when dropParams removes something else', () => {
+    expect(
+      load({ baseURL: 'https://openrouter.ai/api/v1', dropParams: ['temperature'] })
+        ?.reasoningFormat,
+    ).toBe(ReasoningParameterFormat.reasoningEffort);
+  });
+
+  it('does not infer reasoning from the endpoint name alone', () => {
+    expect(load({ name: 'Gateway', baseURL: 'http://localhost:8080/v1', iconURL: 'xai' })).toBe(
+      undefined,
+    );
+  });
+
+  it.each(['https://api.mistral.ai/v1', 'https://api.groq.com/openai/v1'])(
+    'does not infer reasoning for the host %s without verified effort support',
+    (baseURL) => {
+      expect(load({ baseURL })).toBeUndefined();
+    },
+  );
+
+  it('keeps an explicit reasoning format', () => {
+    expect(
+      load({
+        baseURL: 'https://openrouter.ai/api/v1',
+        customParams: { reasoningFormat: ReasoningParameterFormat.disabled },
+      })?.reasoningFormat,
+    ).toBe(ReasoningParameterFormat.disabled);
+  });
+
+  it('defers to explicit reasoning parameter definitions', () => {
+    const customParams = load({
+      baseURL: 'https://openrouter.ai/api/v1',
+      customParams: { paramDefinitions: [{ key: 'thinkingLevel' }] },
+    });
+    expect(customParams?.reasoningFormat).toBeUndefined();
+    expect(customParams?.paramDefinitions).toEqual([{ key: 'thinkingLevel' }]);
+  });
+
+  it('defers to an explicit non-default params endpoint', () => {
+    expect(
+      load({
+        baseURL: 'https://openrouter.ai/api/v1',
+        customParams: { defaultParamsEndpoint: EModelEndpoint.anthropic },
+      })?.reasoningFormat,
+    ).toBeUndefined();
+  });
+
+  it('leaves a native provider endpoint to its own param set', () => {
+    expect(
+      load({ baseURL: 'https://openrouter.ai/api/v1', provider: EModelEndpoint.anthropic })
+        ?.reasoningFormat,
+    ).toBeUndefined();
+  });
+});
+
 describe('loadCustomEndpointsConfig – user credential prompts', () => {
   it('requires a user key when the custom base URL is user-provided', () => {
     const config = loadCustomEndpointsConfig([
diff --git a/packages/api/src/endpoints/custom/config.ts b/packages/api/src/endpoints/custom/config.ts
index 2ebc495086c..19cc4a2e7a9 100644
--- a/packages/api/src/endpoints/custom/config.ts
+++ b/packages/api/src/endpoints/custom/config.ts
@@ -1,9 +1,76 @@
-import { EModelEndpoint, extractEnvVariable, normalizeEndpointName } from 'librechat-data-provider';
+import {
+  ProviderId,
+  EModelEndpoint,
+  extractEnvVariable,
+  normalizeEndpointName,
+  reasoningSettingKeys,
+  ReasoningParameterFormat,
+} from 'librechat-data-provider';
 import type { TCustomEndpoints, TEndpoint } from 'librechat-data-provider';
 import type { TCustomEndpointsConfig } from '~/types/endpoints';
-import { resolveEndpointProviderId } from './providers';
+import { resolveEndpointProviderId, providerFromBaseURL } from './providers';
 import { isUserProvided } from '~/utils';
 
+/**
+ * Hosts that accept an effort for any model they serve. OpenRouter takes it as
+ * `reasoning.effort`, which the OpenRouter request path builds itself. Matched on
+ * the base URL host only: a name or icon says nothing about what the server
+ * behind it accepts.
+ *
+ * OpenAI and xAI are excluded on purpose. Their support and allowed values depend
+ * on the model (gpt-4o takes no effort, grok-4.6 and grok-4.7 take different
+ * sets, GPT-5.6 with tools needs the Responses route a custom endpoint does not
+ * get), and this inference is made per endpoint, before a model is chosen.
+ */
+const effortReasoningHosts: ReadonlySet<ProviderId> = new Set([ProviderId.openrouter]);
+
+/** Backend params whose removal also removes the effort this inference would advertise. */
+const reasoningDropParams: ReadonlySet<string> = new Set(['reasoning_effort', 'reasoning']);
+
+type CustomParams = NonNullable<TEndpoint['customParams']>;
+
+/**
+ * Declares reasoning support for a known host so the effort control appears
+ * without per-endpoint config. Anything the admin stated wins: a native
+ * `provider`, a non-default `defaultParamsEndpoint`, a `reasoningFormat`
+ * (including `disabled`), or reasoning parameter definitions. An endpoint that
+ * drops the effort before sending it is not advertised as supporting it.
+ */
+function withHostReasoning(
+  customParams: TEndpoint['customParams'],
+  baseURL: string,
+  provider?: string,
+  dropParams?: string[],
+): TEndpoint['customParams'] {
+  const host = providerFromBaseURL(baseURL);
+  if (
+    provider != null ||
+    host == null ||
+    !effortReasoningHosts.has(host) ||
+    dropParams?.some((param) => reasoningDropParams.has(param)) === true
+  ) {
+    return customParams;
+  }
+  const params = (customParams ?? {}) as Partial<CustomParams>;
+  const declaresReasoning = params.paramDefinitions?.some((setting) =>
+    reasoningSettingKeys.some((key) => key === setting.key),
+  );
+  const paramsEndpoint = params.defaultParamsEndpoint;
+  if (
+    params.reasoningFormat != null ||
+    declaresReasoning === true ||
+    (paramsEndpoint != null &&
+      paramsEndpoint !== EModelEndpoint.custom &&
+      paramsEndpoint !== ProviderId.openrouter)
+  ) {
+    return customParams;
+  }
+  return {
+    ...params,
+    reasoningFormat: ReasoningParameterFormat.reasoningEffort,
+  } as TEndpoint['customParams'];
+}
+
 /**
  * Load config endpoints from the cached configuration object
  * @param customEndpointsConfig - The configuration object
@@ -37,6 +104,7 @@ export function loadCustomEndpointsConfig(
         modelDisplayLabel,
         customParams,
         provider,
+        dropParams,
       } = endpoint;
       const name = normalizeEndpointName(configName);
 
@@ -55,7 +123,7 @@ export function loadCustomEndpointsConfig(
         (customParams?.defaultParamsEndpoint == null ||
           customParams.defaultParamsEndpoint === EModelEndpoint.custom)
           ? { ...customParams, defaultParamsEndpoint: provider }
-          : customParams;
+          : withHostReasoning(customParams, resolvedBaseURL, provider, dropParams);
 
       customEndpointsConfig[name] = {
         type: EModelEndpoint.custom,
diff --git a/packages/api/src/endpoints/custom/providers.ts b/packages/api/src/endpoints/custom/providers.ts
index 1fed231e75b..a7491a2fb7a 100644
--- a/packages/api/src/endpoints/custom/providers.ts
+++ b/packages/api/src/endpoints/custom/providers.ts
@@ -35,7 +35,7 @@ export const providerHosts: ReadonlyArray<readonly [string, ProviderId]> = [
  * Ollama run on operator-defined hosts. They resolve by iconURL, provider or name.
  */
 
-function providerFromBaseURL(baseURL?: string): ProviderId | undefined {
+export function providerFromBaseURL(baseURL?: string): ProviderId | undefined {
   if (!baseURL) {
     return undefined;
   }
diff --git a/packages/data-provider/src/parameterSettings.ts b/packages/data-provider/src/parameterSettings.ts
index 7d21c90d8ce..1fd98abeb53 100644
--- a/packages/data-provider/src/parameterSettings.ts
+++ b/packages/data-provider/src/parameterSettings.ts
@@ -1322,7 +1322,7 @@ export const agentParamSettings: Record<string, SettingsConfiguration | undefine
   return acc;
 }, {});
 
-const reasoningSettingKeys = [
+export const reasoningSettingKeys = [
   'reasoning_effort',
   'effort',
   'thinkingLevel',