diff --git a/src/sdk/client.ts b/src/sdk/client.ts
index 93e0d3ed..7badfea7 100644
--- a/src/sdk/client.ts
+++ b/src/sdk/client.ts
@@ -2,6 +2,7 @@ import { loadConfig } from "../config/loader";
 import { Config } from "../config/schema";
 import { request as requestClient, requestJson as requestJsonClient, RequestOpts } from "../client/http";
 import { parseSSE } from "../client/stream";
+import { mapApiError, type ApiErrorBody } from "../errors/api";
 import { SDKError } from "../errors/base";
 import { ExitCode } from "../errors/codes";
 import { MiniMaxSDKOptions } from "./types";
@@ -38,14 +39,20 @@ export class Client {
     for await (const event of parseSSE(res)) {
       if (event.data === '[DONE]') break;
       if (!event.data) continue;
+      let chunk: T & ApiErrorBody;
       try {
-        yield JSON.parse(event.data) as T;
+        chunk = JSON.parse(event.data) as T & ApiErrorBody;
       } catch (err) {
         throw new SDKError(
           `Failed to parse stream chunk: ${err instanceof Error ? err.message : String(err)}`,
           ExitCode.GENERAL,
         );
       }
+      const statusCode = chunk?.base_resp?.status_code;
+      if ((statusCode !== undefined && statusCode !== 0) || chunk?.error) {
+        throw mapApiError(res.status, chunk, res.url || undefined);
+      }
+      yield chunk as T;
     }
   }
 }
diff --git a/test/sdk/stream-errors.test.ts b/test/sdk/stream-errors.test.ts
new file mode 100644
index 00000000..79c16ec7
--- /dev/null
+++ b/test/sdk/stream-errors.test.ts
@@ -0,0 +1,88 @@
+import { afterEach, describe, expect, it } from 'bun:test';
+import { MiniMaxSDK } from '../../src/sdk';
+import { CLIError } from '../../src/errors/base';
+import { ExitCode } from '../../src/errors/codes';
+import { createMockServer, sseResponse, type MockServer } from '../helpers/mock-server';
+
+describe('SDK SSE API errors', () => {
+  let server: MockServer;
+
+  afterEach(() => server?.close());
+
+  for (const modality of ['text', 'speech', 'transcription'] as const) {
+    it(`rejects a ${modality} API error without yielding it or subsequent events`, async () => {
+      const first = modality === 'text'
+        ? { type: 'content_block_delta', index: 0, delta: { type: 'text_delta', text: 'Partial' } }
+        : modality === 'speech'
+          ? { data: { audio: 'abcd', status: 1 }, base_resp: { status_code: 0, status_msg: 'success' } }
+          : { delta: 'Partial', index: 0, finish: false };
+      const error = modality === 'text'
+        ? { type: 'error', error: { type: 'insufficient_balance_error', message: 'Balance exhausted' } }
+        : { base_resp: { status_code: 1028, status_msg: 'Insufficient quota' } };
+      server = createMockServer({
+        routes: {
+          '/': () => sseResponse([first, error, first].map(data => ({ data: JSON.stringify(data) }))),
+        },
+      });
+      const sdk = new MiniMaxSDK({ apiKey: 'test-key', region: 'global', baseUrl: server.url });
+      const stream = modality === 'text'
+        ? await sdk.text.chat({ messages: [{ role: 'user', content: 'Hello' }], stream: true })
+        : modality === 'speech'
+          ? await sdk.speech.synthesize({ text: 'Hello', stream: true })
+          : await sdk.speech.transcribe({ file: new Blob(['audio']), stream: true });
+
+      const initial = await stream.next();
+      expect(initial.done).toBe(false);
+      expect(initial.value as unknown).toEqual(first);
+      let caught: unknown;
+      try {
+        await stream.next();
+      } catch (err) {
+        caught = err;
+      }
+      expect(caught).toBeInstanceOf(CLIError);
+      if (modality === 'speech') {
+        expect((caught as CLIError).hint).toContain('Speech models require the Plus plan');
+      }
+      expect(caught).toMatchObject({
+        exitCode: ExitCode.QUOTA,
+        message: expect.stringContaining(modality === 'text' ? 'Balance exhausted' : 'Insufficient quota'),
+        hint: expect.stringContaining(modality === 'text' ? 'billing' : 'mmx quota show'),
+      });
+      expect(await stream.next()).toEqual({ value: undefined, done: true });
+    });
+  }
+
+  it('preserves the content-filter classification of the first speech event', async () => {
+    server = createMockServer({
+      routes: {
+        '/v1/t2a_v2': () => sseResponse([{
+          data: JSON.stringify({ base_resp: { status_code: 1027, status_msg: 'output new_sensitive' } }),
+        }]),
+      },
+    });
+    const sdk = new MiniMaxSDK({ apiKey: 'test-key', region: 'global', baseUrl: server.url });
+    const stream = await sdk.speech.synthesize({ text: 'Hello', stream: true });
+    await expect(stream.next()).rejects.toMatchObject({
+      exitCode: ExitCode.CONTENT_FILTER,
+      message: expect.stringContaining('output new_sensitive'),
+    });
+  });
+
+  it('preserves the message of an Anthropic-compatible server error', async () => {
+    server = createMockServer({
+      routes: {
+        '/anthropic/v1/messages': () => new Response(
+          'event: error\ndata: {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}\n\n',
+          { headers: { 'Content-Type': 'text/event-stream' } },
+        ),
+      },
+    });
+    const sdk = new MiniMaxSDK({ apiKey: 'test-key', region: 'global', baseUrl: server.url });
+    const stream = await sdk.text.chat({ messages: [{ role: 'user', content: 'Hello' }], stream: true });
+    await expect(stream.next()).rejects.toMatchObject({
+      exitCode: ExitCode.GENERAL,
+      message: 'API error: Overloaded (HTTP 200)',
+    });
+  });
+});