diff --git a/src/core/ai/gateway.ts b/src/core/ai/gateway.ts
index fdcae8779..cdd04509c 100644
--- a/src/core/ai/gateway.ts
+++ b/src/core/ai/gateway.ts
@@ -377,7 +377,11 @@ export function applyOpenAICompatConfig(
   cfg: AIGatewayConfig,
 ): { baseURL: string; fetch?: typeof fetch } {
   if (recipe.resolveOpenAICompatConfig) {
-    return recipe.resolveOpenAICompatConfig(cfg.env);
+    const compat = recipe.resolveOpenAICompatConfig(cfg.env);
+    return {
+      ...compat,
+      fetch: compat.fetch ?? recipe.compat?.fetch,
+    };
   }
   const baseURL = cfg.base_urls?.[recipe.id] ?? recipe.base_url_default;
   if (!baseURL) {
@@ -386,7 +390,10 @@ export function applyOpenAICompatConfig(
       recipe.setup_hint,
     );
   }
-  return { baseURL };
+  return {
+    baseURL,
+    fetch: recipe.compat?.fetch,
+  };
 }
 
 /**
diff --git a/src/core/ai/recipes/deepseek.ts b/src/core/ai/recipes/deepseek.ts
index daa7cf2e9..a8ae6388f 100644
--- a/src/core/ai/recipes/deepseek.ts
+++ b/src/core/ai/recipes/deepseek.ts
@@ -1,5 +1,57 @@
 import type { Recipe } from '../types.ts';
 
+/**
+ * DeepSeek-style reasoning models can return the assistant answer in
+ * `message.reasoning_content` while leaving `message.content` empty. The AI
+ * SDK's OpenAI-compatible adapter reads only `content`, so promote the field
+ * before the SDK parses the response.
+ */
+export const deepseekReasoningContentCompatFetch = (async (
+  input: RequestInfo | URL,
+  init?: RequestInit,
+) => {
+  const resp = await fetch(input, init);
+  if (!resp.ok) return resp;
+  const ct = resp.headers.get('content-type') ?? '';
+  if (!ct.toLowerCase().includes('application/json')) return resp;
+
+  try {
+    const json: any = await resp.clone().json();
+    if (!json || typeof json !== 'object' || !Array.isArray(json.choices)) {
+      return resp;
+    }
+
+    let modified = false;
+    for (const choice of json.choices) {
+      const message = choice?.message;
+      if (!message || typeof message !== 'object') continue;
+      const content = message.content;
+      const reasoningContent = message.reasoning_content;
+      const contentIsEmpty =
+        content === null ||
+        content === undefined ||
+        (typeof content === 'string' && content.trim().length === 0);
+      if (
+        contentIsEmpty &&
+        typeof reasoningContent === 'string' &&
+        reasoningContent.trim().length > 0
+      ) {
+        message.content = reasoningContent;
+        modified = true;
+      }
+    }
+
+    if (!modified) return resp;
+    return new Response(JSON.stringify(json), {
+      status: resp.status,
+      statusText: resp.statusText,
+      headers: resp.headers,
+    });
+  } catch {
+    return resp;
+  }
+}) as unknown as typeof fetch;
+
 /**
  * DeepSeek exposes an OpenAI-compatible /v1/chat/completions endpoint.
  * Useful as the second hop in a refusal-fallback chain and for cheap-
@@ -16,6 +68,9 @@ export const deepseek: Recipe = {
     required: ['DEEPSEEK_API_KEY'],
     setup_url: 'https://platform.deepseek.com/api_keys',
   },
+  compat: {
+    fetch: deepseekReasoningContentCompatFetch,
+  },
   touchpoints: {
     chat: {
       models: ['deepseek-chat', 'deepseek-reasoner'],
diff --git a/src/core/ai/types.ts b/src/core/ai/types.ts
index 4dced5c6d..6601cb828 100644
--- a/src/core/ai/types.ts
+++ b/src/core/ai/types.ts
@@ -326,6 +326,15 @@ export interface Recipe {
     baseURL: string;
     fetch?: typeof fetch;
   };
+  /**
+   * Optional OpenAI-compatible transport shims for recipe-specific wire quirks.
+   * Kept separate from `resolveOpenAICompatConfig()` so a recipe can install a
+   * static fetch wrapper while preserving the normal `base_urls[recipe.id]`
+   * override chain.
+   */
+  compat?: {
+    fetch?: typeof fetch;
+  };
   /**
    * v0.32 (D13=A): optional runtime readiness check for local-server
    * recipes (ollama, llama-server, future lmstudio-recipe). Returns
diff --git a/test/deepseek-reasoning-content.test.ts b/test/deepseek-reasoning-content.test.ts
new file mode 100644
index 000000000..dc211b8e8
--- /dev/null
+++ b/test/deepseek-reasoning-content.test.ts
@@ -0,0 +1,80 @@
+import { afterEach, describe, expect, test } from 'bun:test';
+import {
+  chat,
+  configureGateway,
+  resetGateway,
+} from '../src/core/ai/gateway.ts';
+
+const originalFetch = globalThis.fetch;
+
+function installDeepSeekResponse(message: Record<string, unknown>): void {
+  globalThis.fetch = (async () => {
+    const json = {
+      id: 'fake-deepseek-chatcmpl',
+      object: 'chat.completion',
+      created: 0,
+      model: 'deepseek-reasoner',
+      choices: [
+        {
+          index: 0,
+          message: {
+            role: 'assistant',
+            ...message,
+          },
+          finish_reason: 'stop',
+        },
+      ],
+      usage: { prompt_tokens: 3, completion_tokens: 5, total_tokens: 8 },
+    };
+    return new Response(JSON.stringify(json), {
+      status: 200,
+      headers: { 'content-type': 'application/json' },
+    });
+  }) as unknown as typeof fetch;
+}
+
+async function runDeepSeekChat(): Promise<string> {
+  configureGateway({
+    chat_model: 'deepseek:deepseek-reasoner',
+    env: { DEEPSEEK_API_KEY: 'sk-deepseek-fake' },
+  });
+  const result = await chat({
+    model: 'deepseek:deepseek-reasoner',
+    messages: [{ role: 'user', content: 'summarize this chunk' }],
+  });
+  return result.text;
+}
+
+afterEach(() => {
+  globalThis.fetch = originalFetch;
+  resetGateway();
+});
+
+describe('DeepSeek reasoning_content compatibility fetch', () => {
+  test('empty content + reasoning_content promotes reasoning text into chat output', async () => {
+    installDeepSeekResponse({
+      content: '',
+      reasoning_content: 'The synopsis lives here.',
+    });
+
+    await expect(runDeepSeekChat()).resolves.toBe('The synopsis lives here.');
+  });
+
+  test('non-empty content + reasoning_content passes content through unchanged', async () => {
+    installDeepSeekResponse({
+      content: 'Use the final answer.',
+      reasoning_content: 'Do not duplicate this reasoning.',
+    });
+
+    await expect(runDeepSeekChat()).resolves.toBe('Use the final answer.');
+  });
+
+  test('empty content + empty reasoning_content stays empty without crashing', async () => {
+    installDeepSeekResponse({
+      content: '',
+      reasoning_content: '',
+    });
+
+    await expect(runDeepSeekChat()).resolves.toBe('');
+  });
+});