Gemini-omni-flash-preview started returning PERMISSION_DENIED after previously working

Gemini API
1

Hello Google Gemini API team,

I have a production integration using Gemini Omni Flash Preview through the Gemini Interactions API that was working successfully and suddenly started returning PERMISSION_DENIED.

Environment:

Model:
gemini-omni-flash-preview

API:
Gemini Developer API / Interactions API

Endpoint:
POST /v1beta/interactions

Project:
Smart Corretor Ai

Billing:
Paid Tier 2 / prepaid, with positive balance

LAST CONFIRMED SUCCESS

On August 10, 2026, this exact integration completed successfully.

The interaction processed for approximately 59 seconds and returned the final MP4 normally.

CURRENT FAILURE

Since August 13–14, the same flow consistently fails.

The sequence is:

  1. POST /v1beta/interactions succeeds.
  2. Google creates and returns a valid interaction_id.
  3. Approximately 5–7 seconds later, the first status/SSE lookup returns:

event_type = error
error.code = permission_denied

Message:

“There was a problem processing your request. You will not be charged.”

Google AI Studio usage monitoring also records the failed requests as 403 Forbidden.

LATEST REPRODUCTION

Date:
August 14, 2026

The interaction was created successfully and failed approximately 6.4 seconds later with:

event_type=error
code=permission_denied

No video was generated and Google reported that the request would not be charged.

WHAT WE HAVE ALREADY VERIFIED

  • Billing is active.
  • Prepaid balance is positive.
  • Project is Paid Tier 2.
  • Gemini Omni Flash appears available in AI Studio.
  • RPM/RPD quotas are not exhausted.
  • The original API key that worked on August 10 was tested.
  • A brand-new Google AI Studio authorization key was also tested.
  • The new key belongs to the same project, is bound to a service account, and is explicitly restricted to Gemini API.
  • Both keys produce the same permission_denied result.
  • Other Google video generation workloads in the same billing project continue to work.
  • We restored our application to the same Gemini Omni request contract used by the successful August 10 job.
  • Same model ID.
  • Same /v1beta/interactions endpoint.
  • Same generation payload structure.
  • No Vertex AI is involved.

The important detail is that authentication and interaction creation succeed. The PERMISSION_DENIED happens only after the interaction has already been created, when asynchronous processing begins.

QUESTIONS FOR THE GEMINI TEAM

  1. Has access, entitlement, allowlisting, quota policy, or availability for gemini-omni-flash-preview changed between August 10 and August 14?

  2. Can Google check whether our project still has the required entitlement for Gemini Omni Flash Preview background video generation?

  3. Why would interaction creation succeed but asynchronous processing return permission_denied a few seconds later?

  4. Is there an internal reason/status associated with these failed interactions that is not exposed through the public SSE error?

  5. Is any project-side remediation required?

I can provide the full interaction IDs privately to Google staff if needed.

I would prefer not to change models because this exact integration was working successfully a few days ago.

Thank you.