Since I’m always coming here with issues, I figured it’s only fair to come here with good news as well.
My last post I was really upset with how my quota was being eaten alive on Gemini Flash’s Lowest model, my quota just felt like it was being robbed from me every week, but that’s changed!
I am super happy with Gemini Flash’s Low model now, I’m completing an entire day’s work using only 20% of my weekly quota!
So big thank you to Google for relieving a ton of stress and fear. I was really losing hope there for a while.
Claude continues to be extremely expensive for what it offers, I blow through my entire Claude quota in 2 days, so cheers to you Google.
Concurring. I’m finding the 3.5 flash models … They do require firm guidance compared to the Claude models. What has worked well for me across a few iterations the past few weeks is:
I begin by writing down the idea (intent, feature, etc.) in a working document (markdown).
I iterate on the document with Claude, either Sonnet or Opus depending on the complexity, until I have a plan for implementation.
I get a fresh Gemini Flash 3.5 (High) agent to review the design and implementation plan, thoroughly and adversarially, and leave comprehensive feedback in a working-doc-review.md file (strict orders not to edit the working-doc.md.)
I re-engage Claude (fresh session or, usually, back to the session I had going earlier), and indicate the feedback to it with instructions to evaluate the whole design in light of the review, make sure to discriminate what makes sense and what doesn’t.
I might do 3 & 4 iteratively.
I get Gemini to implement the whole thing.
This way I can crank out 3k - 10k lines of solid code (I work in C++) in one 5h window without burning out either Claude or Gemini quota for that window. Claude does well with the research work, GF3.5 does exceedingly well with the heavy lifting of writing a ton of code fitting into a (relatively) huge pile of context. Gemini isn’t bad at research, but I’m finding it’s much more useful for review than for initial design.
(I use a specific project-workflow skill that I’ve been honing for a while, lmk if I should post it.)
It works reasonably well if you simply want to create a good-looking mock-up, but performance suffers significantly in every other respect. How do you all deal with this issue? If anyone has any insights, please share your advice.
I wanted to share my feedback regarding the evaluation experience of Antigravity. I started testing the tool using Gemini on the free tier to assess its viability for a future investment. Honestly, the quota limitation is impressively ridiculous.
It completely breaks the workflow and makes it impossible to properly test the tool’s capabilities or see if it is worth paying for. Strong competitors in the market—like Codex and many others—do not impose such restrictive barriers during the evaluation phase. They actually allow users to understand the tool’s value before asking for a credit card.
Because of this, I have completely given up on using Antigravity. I truly hope the team heavily reconsiders this approach if you want to keep attracting developers and businesses to your ecosystem.
Google’s keep changing the quota and policy behind the scene that I have lost faith in them…
I’m currently subbed to Claude (for planning) and GPT 5.5 for implementation…
it was Gemini for the implementation, back when their quota was true to their claim…
As a “Pro” package user who has been developing projects with Antigravity for months, I am far more frustrated by how quickly my quota runs out than by the discontinuation of the “Unlimited” plan. The last time I saw my quota depleted so rapidly was due to a bug following an update back when Antigravity 1.0 was in use. If this move is an attempt to cut costs. I refuse to upgrade to a higher-tier package!