Gemini Live API WebSocket Error 1008: "Operation is not implemented, or supported, or enabled"

[Adhishtanaka]

WebSocket connection to Gemini Live API closes unexpectedly with error code 1008 and reason: "Operation is not implemented, or supported, or enabled."

Environment

Component	Version/Value
SDK	@google/genai (latest)
Model	models/gemini-2.5-flash-native-audio-preview-12-2025
Platform	Browser (Chrome/Edge) via Next.js 16.x
Connection Type	Direct WebSocket (Live API)
Audio Input	16-bit PCM, 16kHz, mono, little-endian
Audio Output	16-bit PCM, 24kHz, mono, little-endian

---

Error Details

Console Output

direct-gemini-client.ts:244 🔌 Gemini connection closed (code: 1008, reason: Operation is not implemented, or supported, or enabled.)

Steps to Reproduce

1. Initialize GoogleGenAI client with valid API key
2. Connect to Live API with audio modality:

import { GoogleGenAI, LiveConnectConfig, Modality } from "@google/genai";

const client = new GoogleGenAI({ apiKey: "VALID_API_KEY" });

const config: LiveConnectConfig = {
  responseModalities: [Modality.AUDIO],
  speechConfig: {
    voiceConfig: {
      prebuiltVoiceConfig: {
        voiceName: "Aoede"
      }
    }
  },
  systemInstruction: {
    parts: [{ text: "You are a helpful assistant." }]
  },
  // Generation config fields set directly (not nested)
  temperature: 0.7,
  topP: 0.9,
  topK: 40,
  // Realtime input config
  realtimeInputConfig: {
    automaticActivityDetection: {
      disabled: true
    }
  }
};

await client.live.connect({
  model: "models/gemini-2.5-flash-native-audio-preview-12-2025",
  config,
  callbacks: {
    onopen: () => console.log("Connected"),
    onclose: (event) => console.log(`Closed: ${event.code} - ${event.reason}`),
    onerror: (error) => console.error("Error:", error),
    onmessage: (message) => console.log("Message:", message)
  }
});

3. Start sending audio chunks (16-bit PCM, 16kHz):

session.sendRealtimeInput({
  media: {
    mimeType: "audio/pcm;rate=16000",
    data: base64AudioChunk // 20-40ms chunks as recommended
  }
});

4. Connection closes with error 1008 after a few seconds or during first audio response

---

Expected Behavior

• WebSocket connection should remain stable
• Audio input should be processed
• Audio response should be received completely
• Connection should only close on explicit disconnect or timeout

---

Actual Behavior

• Connection establishes successfully (onopen fires)
• First audio response may start streaming
• Connection abruptly closes with code 1008
• Error message: "Operation is not implemented, or supported, or enabled."

---

Deprecation Warning Observed

Before the error, this deprecation warning appears:

Setting `LiveConnectConfig.generation_config` is deprecated, please set the fields on `LiveConnectConfig` directly. This will become an error in a future version (not before Q3 2025).

Note: Even after fixing the deprecated config (moving generationConfig fields to top-level), the 1008 error still occurs.

---

Configuration Attempts

Attempt 1: Original (with deprecated nested config)

// ❌ Deprecated - causes warning
const config = {
  responseModalities: [Modality.AUDIO],
  generationConfig: {
    temperature: 0.7,
    topP: 0.9,
    topK: 40
  }
};

Attempt 2: Fixed (fields at top level)

// ✅ Correct per deprecation notice - still causes 1008
const config = {
  responseModalities: [Modality.AUDIO],
  temperature: 0.7,
  topP: 0.9,
  topK: 40
};

Attempt 3: With realtimeInputConfig

// Tried disabling automatic activity detection - still causes 1008
const config = {
  responseModalities: [Modality.AUDIO],
  temperature: 0.7,
  realtimeInputConfig: {
    automaticActivityDetection: {
      disabled: true
    }
  }
};

---

Audio Streaming Details

Following Google's best practices for streaming:

Setting	Value	Best Practice
Chunk Size	20-40ms	✅ Recommended
Sample Rate (Input)	16kHz	✅ Required
Sample Rate (Output)	24kHz	✅ Expected
Bit Depth	16-bit PCM	✅ Required
Channels	Mono	✅ Required
Endianness	Little-endian	✅ Required

---

Workarounds Attempted

• Moved generationConfig to top-level fields
• Added realtimeInputConfig.automaticActivityDetection.disabled: true
• Reduced audio chunk size to 20-40ms
• Validated audio format (16-bit PCM, 16kHz, little-endian)
• Added connection timeout handling
• Implemented reconnection logic
• None resolved the 1008 error

Request

Please provide:

1. Root cause of the 1008 error
2. Correct configuration for stable Live API WebSocket connections
3. Any model-specific requirements for audio streaming
4. Updated documentation if the API behavior has changed

---

Additional Context

This issue occurs in a real-time voice interview application where:

• User speaks → Audio captured at 16kHz
• Audio sent to Gemini Live API in small chunks (20-40ms)
• AI responds with audio at 24kHz
• Response played back to user

The application works correctly for initial connection but fails intermittently with 1008 errors, making it impossible to complete a full voice conversation.

[shmishra99]

@Adhishtanaka , Are you experiencing this issue only with the gemini-2.5-flash-native-audio-preview-12-2025 model, or does it happen with the 09-2025 version as well?

[Adhishtanaka]

well in 09-2025 it working but in 12 its very unstable and shows this 1008 error most of the time

[shmishra99]

@Adhishtanaka , This appears to be a model-specific issue. I have created an internal bug to investigate issue.

[Adhishtanaka]

@Adhishtanaka , This appears to be a model-specific issue. I have created an internal bug to investigate issue.

Thanks

[iAmTheAbdullah]

Hi,

I am experiencing the exact same problem where I get intermittent 1008 “Operation is not implemented, or supported, or enabled” error at very random times with gemini-2.5-flash-native-audio-preview-12-2025

I've seen on multiple forums (example below) that this is still happening.

Forum -> https://discuss.ai.google.dev/t/gemini-live-api-websocket-error-1008-operation-is-not-implemented-or-supported-or-enabled/114644/45

Is there an update for the fix of this issue?

[the-kool-sk]

For me this happens only when user asks any recent news question. For which model might need to look up to google search. For normal conversations it works flawlessly.

[Adhishtanaka]

In my case, the issue was resolved when I switched to gemini-2.5-flash-native-audio-preview-09-2025 in the early days. However, I stopped using that model because my supervisor advised me not to use preview APIs in production. So I avoided it for a while.

This week, I tried using it again, and it worked. Both the 09 and 12 versions are working for me now, including tool calling. However, I noticed that although it works in my case, many others are still facing the same issue, especially when using tool calling.

It also appears that the Google team is actively working on this problem, as mentioned in the discussion thread below:

https://discuss.ai.google.dev/t/gemini-live-api-websocket-error-1008-operation-is-not-implemented-or-supported-or-enabled/114644/45

We may need to wait for the next stable model release for a complete fix.i guess