vercel/ai · error · Error

Google realtime auth token request failed: ${response.status

Error message

Google realtime auth token request failed: ${response.status} ${text}

What it means

After POSTing to Google's v1alpha auth_tokens endpoint to create a realtime client secret, the model checks response.ok; on any non-2xx it reads the response body and throws this error including the HTTP status and Google's error text. It surfaces upstream auth/token-creation failures (bad key, invalid session config, quota, unsupported model) with Google's own message for diagnosis.

Source

Thrown at packages/google/src/realtime/google-realtime-model.ts:103

      `${getAuthTokensURL(this.config.baseURL)}?key=${encodeURIComponent(apiKey)}`,
      {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({
          // `uses: 0` means no limit is applied to how many times the token can
          // start a session (per the AuthToken spec). An unset value would
          // default to 1, which breaks WebSocket reconnects within the session.
          uses: 0,
          expireTime,
          newSessionExpireTime,
          bidiGenerateContentSetup: setupPayload,
        }),
      },
    );

    if (!response.ok) {
      const text = await response.text();
      throw new Error(
        `Google realtime auth token request failed: ${response.status} ${text}`,
      );
    }

    const data = (await response.json()) as {
      name: string;
      expireTime?: string;
    };

    return {
      token: data.name,
      url: getWebSocketURL(this.config.baseURL),
      expiresAt: data.expireTime
        ? Math.floor(new Date(data.expireTime).getTime() / 1000)
        : undefined,
    };
  }

View on GitHub (pinned to 69428b1f8b)

Solutions

1. Read the status and text in the error message: 401/403 means key or permission issues, 400 means bad session config or model, 429 means quota.
2. Verify the API key is valid and the Generative Language / Live API is enabled for the project.
3. Validate the sessionConfig (model id, modalities, tools) against the Live API requirements.
4. Retry with backoff only for 429/5xx; fix config for 400/401/403.

Example fix

// before
const secret = await model.createClientSecret({ sessionConfig: { model: 'made-up-model' } });
// after
const secret = await model.createClientSecret({
  sessionConfig: { model: 'gemini-2.5-flash-native-audio-preview-09-2025' },
});

try {
  const secret = await model.createClientSecret(options);
} catch (error) {
  if (error instanceof Error && error.message.includes('auth token request failed')) {
    const status = error.message.match(/failed: (\d{3})/)?.[1];
    if (status === '429') await backoffAndRetry();
    else logConfigError(error.message); // 400/401/403 need fixes, not retries
  } else { throw error; }
}

When it happens

Trigger: Creating a realtime client secret when Google rejects the auth_tokens request: invalid or revoked API key, malformed bidiGenerateContentSetup session config, referencing an unsupported model, project quota exhausted, or the API not enabled for the project.

Common situations: Wrong-project API key without the Live API enabled; invalid modelId in the session setup; expired/rotated key; regional endpoint misconfiguration; 429 from rate limits.

Related errors

• Cartesia realtime client secret request failed: ${response.s
• Google Generative AI API key is required for realtime token
• Cartesia realtime client secret response did not include a t
• Google Generative AI API key is required for streaming trans
• ${readErrorMessage({ value, status: response.status })}

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/df38f4ac5245bc03. Report an issue: GitHub.