After more than three months of using Google Antigravity, I’ve decided to cancel my subscription. Overall, it has been a frustrating experience.
The biggest issue is the constant usage limits. Even on a paid plan, you hit restrictions far too quickly—especially during real development work. This disrupts workflow and makes the tool unreliable for professional use. It’s simply not something you can depend on when you actually need it.
Performance is also inconsistent. At times, the output is solid, but in many cases it feels shallow or misses key context. This often results in spending more time fixing responses than the time it saves.
For a paid product, I expected stability, predictability, and clear value for money. Instead, it feels like you’re constantly working around system limitations.
Overall, in its current state, it’s not worth the subscription. I’ve moved on to look for more reliable alternatives.
you don’t have the google badge so it means you’re not working for google. Obviously no google employee would ever offer a refund without knowing the details first, but yea, this is reportable stuff
I just experienced this yesterday (along with numerous other times) and was able to continue working once I selected a different model. It was discouraging because I felt I had originally selected the correct language model for the work, but soon discovered that the model switch didn’t effect context or conversation history. I moved from Gemini 3.1 Pro to Claude Sonnet 3.5 and Claude was efficiently pulling from my past conversations with Gemini to maintain context and collect necessary info from my previous work. Have you tried using different models after hitting the limit? I almost upgraded to a ridiculously expensive tier just to continue completing tasks, but something told me to try a new model and actually got things done. Hope this helps.
Thanks for sharing your experience. Yes, I tried switching between different models, and in some cases it helped temporarily. The cross-model context sharing is definitely a useful feature.
My main concern, though, is that on a paid plan I shouldn’t have to constantly switch models just to continue working. When you’re deep into real development tasks, changing models can affect output quality, reasoning style, and consistency.
I agree the workaround exists, but for a professional workflow I expected a more stable and predictable experience without constantly managing limits or model hopping.
I agree 100%. This rate limit structure appears to be an “automatic” subscription upsell built into the normal use of Antigravity. Either it’s not suitable for production-ready iteration, or it’s an intentional attempt to simply grow revenue among core users. I’ve adopted a Google-first workflow and this issue could be the reason my firm makes a tech stack change.