Even though I have a Google AI Pro subscription, all my model quotas in the Antigravity app suddenly dropped to zero. Strangely enough, when I ask a question to the Claude Opus 4.6 model without using my available AI credits, I can still get a response, but my quotas still appear as completely empty. Is anyone else experiencing this issue?
same over here - quota 0 on all models even I have used only model Flash for 3 short prompts today (and no usage at all for 4 days before today)
i have not used AG for a while and just entered to generate a single image and i can confirm all quota is 0. nice job google
I have tokens but it never responds to my request
same quota 0 but didnāt use anything in the last days
Same over here. But Iām working with the models (with more interruptions than other days but working)
same, It had been quite a while since Iād had any errors, and today I got the message about the panel being full of exhausted models and itās giving me errors.
It happened last night around midnight when suddenly AntiGravity started giving me a typical āretryā error. When I looked at the modelsā quota, it was all zero. I havenāt even touched Pro models in a week, and they are all zero. I do have a Pro subscription as well. AntiGravity is the most error-prone IDE. It always has some issues with it.
Also there is no email or any statement from google either
I had some āRetryā errors after days without them, but Iāve been working this morning without any important problems, even with 0% quota left and warning icon appearing in āModelsā in settings.
Antigravity is a nest of problems, thatās true. But when it works, I think thereās no rival. The combination AG + Claude Opus 4.6 is simply magnificent (but yes, it works for 1h and after you have to wait hours or days for more quota).
In my job we use ULTRA tier (and we run into the same issues), but Iām considering paying for another Agentic IDE or subscription for my personal usage on weekends.
Do you know any reliable alternative?
The quotas are up and fine, while Claude and GPT-OSS are at 80% usage, with 4 days remaining until refresh, which I havenāt touched in a week.
I have been using CoPilot Pro and AntiGravity Gemini Pro both, as I have been working on multiple projects at a time.
Gemini is reasonable with codebase analysis and preparing tasks. I tried with Claude and GPT, and just one single prompt hit me with 40% usage.
I havenāt tried Codex and other yet (with agent).
Thanks for your reply, mate.
My best experience is with Antigravity and Opus 4.6. No doubt. But now Iām changing the point of my usage. Iām trying Opus 4.6 to make architecture and detailed tasks specification, in order to persist it and feed another smaller or lower model (Gemini 3.1 Pro or Flash). Maybe doing a better specs doc makes the job easier for those.
Exactly what I am doing. I plan with a higher model and implement with a lower model (Gemini Flash), as it will only follow what is in the plan. Secondly, if revisions are needed, I try first with Flash, if it goes crazy (as it did most of the time), then I switch to a higher model. Flash is good, but if you ask it three or more tasks in one prompt, it just does the first one 
I mainly use Antigravity Ultra together with Cloud Code Pro for development, especially for Java, Spring Boot, React, Firebase and AI-related projects.
My workflow is usually multi-agent:
- One agent focuses on architecture and project structure.
- One agent generates or refactors code.
- Another agent reviews/debugs the output.
- Sometimes I use a dedicated agent only for documentation, prompts or test generation.
For coding tasks, I switch models depending on the context:
- Gemini Flash agents for fast iterations and brainstorming.
- More advanced Gemini/Cloud Code models for deeper reasoning, refactoring and large-context analysis.
- ChatGPT/Codex when I need stronger code orchestration, debugging explanations or structured outputs.
I also integrate agents with:
- Git repositories
- Firebase
- Google Cloud
- N8N automations
- AI Studio APIs
- RAG/document-based workflows
The most important part is not just the model quality, but the orchestration between agents:
context sharing, avoiding conflicting refactors, keeping architecture consistent and managing token/context windows correctly.
In practice, I treat AI agents almost like a small software team:
one plans, one implements, one reviews, one documents.