Gemini 4 Argon Is Consuming My Entire Weekly Quota Without Completing a Single Task

I am a Google AI Pro subscriber using Google Antigravity for software development.

My experience with Gemini 4 Argon has been extremely frustrating.

I assign a single development task to Argon. It starts working and launches four sub-agents. However, before the task is completed, my entire available quota is exhausted.

The result? I lose access to Argon for an entire week, and the original task remains unfinished.

This is not a productive experience for a paid subscriber.

I understand that advanced reasoning models require more computational resources. However, users should not have to sacrifice their entire weekly quota for a single incomplete development task.

I would like Google to clarify:

  1. Why does Argon consume so much quota during a single agentic coding task?
  2. How much quota does each sub-agent consume?
  3. Why is there no effective quota protection when an agent cannot complete its assigned task?
  4. Will Google consider refunding quota consumed by unsuccessful or incomplete agent executions?
  5. Does Google plan to increase the Gemini 4 Argon quota for AI Pro subscribers?

I am not asking for unlimited usage. I am asking for a reasonable amount of completed work from a paid subscription.

A powerful model is not useful to developers if its quota runs out before it can finish a single task.

Please address this issue with a concrete explanation rather than a generic response about computational costs.

  1. Will Google consider refunding quota consumed by unsuccessful or incomplete agent executions?

I guess NO.

  1. Does Google plan to increase the Gemini 4 Argon quota for AI Pro subscribers?

It did few weeks ago on Argon release, now we’re getting shrinked plans back again.

  1. Why is there no effective quota protection when an agent cannot complete its assigned task?

It’s not their business. I’ve only seen this generosity in Codex in May’26 where even if quota was completely drained agent never stopped untill finishing the task. Golden age.

Hi @Al_Shimbil_Khan,

Welcome to the Forum, and thank you for sharing this feedback.

To address your questions directly:

  • **Subagent & Argon quota usage **: Per the Plans documentation, quota correlates with total agent work, thinking level, and tokens used. Because each subagent runs as a concurrent session (inherit by default), spawning four subagents from Gemini 4 Argon runs five parallel Argon sessions against your quota. You can stop unneeded subagents via Stop Subagent in the subagent panel or use Low/Medium thinking in the model selector.
  • **Completing your task & quota policy **: Consumed quota isn’t refunded if a limit is reached mid-task, and we have shared your feedback on Argon Pro limits with our team. However, you can finish your task right away in that same conversation by switching the model selector to Gemini 3.8 Flash or Claude 5.5 (which have separate quota pools), or by setting AI Credit Overages to Always in Settings to use AI credits.