Dear community,
Today, I would like to share some insights into building with Google Antigravity. First of all, I am a Google AI Pro user and I am successfully building agentic AI with Antigravity.
I have to admit that even now, you can do it by using Gemini native models and only the anthropic models for real issues and problems.
Now let’s get to the topic. Building with Antigravity used to be really easy and forgiving, as we had generous rate limits, even with the Anthropic models, which are likely hosted by Google as they own part of Anthropic and supply them with their TPUs. However, it’s now really hard to use the Antropic models for complex issues as they have gradually reduced usage. Before that, you could easily use Antropic models as and when you needed them. But especially the new Opus 4.6 is really unusable. I can only use it twice before having to wait 70 hours — during which time I can’t use Opus 4.6, Opus 4.5 or even Sonnet.
The problem is that the Gemini models already lag behind the Antropic models or GPT-5.3-Codex in SWE tasks. For now, it makes more sense to use the 300 credits in GitHub Copilot with GPT-5.3-Codex, which, to be honest, seems even better than Opus 4.6 at the moment.
Perhaps we should use Opus only for research or planning before letting 5.3-Codex take over.
I know it’s not in Google’s interest to let people abuse their limits, but we’re trying to build real things with Antigravity. If we don’t get the diversity of models that we get in Cursor or Copilot, it makes no sense to use Antigravity, even if it is really good.
GitHub Copilot is leaping forward with interactive features like asking and memory creation and we don´t need to start talking about Cursor, OpenCode or Claude Code (which you can use with other models as well, nearly unlimited if you combine it with ur Github Copilot Subscriptions, Nvidia and Openrouter API´s…).
If Google wants developers to use Antigravity above everything else, they need to improve the limits and provide more access to models. Also, I can’t wait for new Gemini models like Gemini 3.5. What’s taking them so long to release models?
