Gemini API — HTTP 400 response for gemini-4-argon video request

I’m investigating an HTTP 400 BadRequest response from the Gemini API.

Project: Microduck
Model: gemini-4-argon
Endpoint: Gemini generateContent API
Request: Video analysis with structured JSON output
Request body: Approximately 1.96 MB
Date: October 14, 2026

Our application successfully established a verified TLS connection and transmitted the request. Google returned HTTP 400, but the client incorrectly discarded the response body.

We have since corrected the client’s error handling and audited the complete request, including 66 structured-output schema nodes, without identifying a confirmed API-contract violation.

Google AI Studio shows an HTTP 400 error associated with project usage, but the Logs and Datasets page contains no historical request record.

Could you help determine:

  1. Whether historical server-side diagnostic details are available for the rejected request.

  2. Whether the model supports the supplied structured-output configuration and inline video input.

  3. Whether the rejection could be related to schema complexity, payload validation, or model-specific restrictions.

  4. Whether there are additional diagnostics available for HTTP 400 responses that occurred before request logging was enabled.

I can provide a sanitized request structure, approximate timestamp, and additional technical details privately if needed.

Thank you.

Hello @SilomonTolomeo,

Yes, Gemini 4 Argon supports inline video input (for total request payloads under 20 MB) together with structured output.

  • The details of a rejected HTTP 400 call are only returned in the error response body (error.message and error.details) at the time of the request. They can’t be retrieved afterwards (see API errors).
  • That call won’t appear in Logs and Datasets. For projects created before October 26, 2026, generateContent logging was off by default, past requests are not backfilled, and requests that contain video are not logged.
  • An HTTP 400 INVALID_ARGUMENT on this kind of request usually comes from either the schema (for example, unsupported JSON Schema keywords or very deep nesting) or the request payload (for example, video bytes that can’t be decoded or invalid parameter values).

If it happens again, could you please share the full error.message and error.details from the response so we can look into it further?