Unpopular opinion - the current Ultra limits are actually great for heavy workloads

Despite the recent criticism - and even my own past posts discussing previous issues - I’d like to share my positive experience, which has been very stable lately.

I have the Ultra subscription, which includes priority access during heavy loads, so the experience on a Pro account might differ.

For at least the past three weeks, I’ve had no loading freezes or “high demand” server errors. It’s been working stably and fast.

Regarding the limits: yes, they have changed. Honestly, I think the initial limits were a bit unrealistic, and freebies can only go so far. The ugly part was changing them without notice - that’s a really weak move for a reputable company. BUT, other than that, I’m pretty happy with the current caps. I heavily use Gemini 3.1 Pro on planning modes, running 2-3 agents simultaneously across two screens. I barely hit the limits, and the model itself is pretty solid.

I don’t fully understand all the criticism here. In my experience, Gemini does amazing in all kinds of coding assignments. Claude handles more complex tasks a bit better but struggles with context overload sooner. I like using both, and I’m thankful to have access to them here, even though we’re still on Opus 4.6 for now.

Many people are demanding higher limits on the Opus models. This is Google Antigravity, not Claude Code. Having a third-party model available here is a pure bonus, and I hope they keep it that way. However, I wouldn’t be amazed if that bonus is gone one day (hopefully not without notice this time).

Overall, my Ultra subscription has allowed me to handle multiple complex projects with the freedom to tackle various side-quests. I’m confident I wouldn’t be able to get this much done with any other subscription out there. Even with the nerfed limits, they are still great.

I also don’t find Ultra too expensive considering the enormous package of benefits it comes with: flow credits with unlimited lazy generations, higher limits across the board, and a massive 30TB of cloud storage.

I think the Google team needs to hear the good feedback too, so they have more motivation to continue developing Antigravity. It genuinely has the potential to be the best platform among them.

I’ve also noticed the reduction in timeout errors over the last two weeks. And I don’t run out of AI credits that fast, which allows me to do a lot of work using Claude Opus. While I’m still disappointed in the overall stability of Antigravity (especially the loss of conversations, and the Terminal crashing, requiring a reboot), it’s still good enough that I’m keeping my Ultra subscription.