If you’ve been tracking Google’s rapid iterations inside Antigravity, you’ll remember that they recently spun up a “Low-effort” variant of Gemini 3.5 Flash. The goal was simple: stop the AI from overthinking basic requests and swallowing your entire token quota on simple coding tasks. However, the execution didn’t quite land perfectly. Google is once again rolling out a new version of Gemini in Antigravity to fix some of the issues that have cropped up since its last iteration.

Varun Mohan, Director at Google DeepMind, working on Antigravity, announced on X that the team has deployed a refreshed version of this Gemini 3.5 Flash model. According to Mohan, this updated variant boasts much less and features significantly higher endurance when tackling harder software engineering tasks.

Gemini 3.5 Flash (Low) cut token generation by roughly 45% compared to the original model (now dubbed the “Medium” variant), giving developers a lighter tool for lightweight tasks. While it stopped the token bleeding on minor jobs, developers noticed a significant drop-off in output quality and structural consistency when a task required just a bit more heavy lifting. Google essentially traded efficiency for endurance, leaving a blind spot when a “simple” task unexpectedly required a deeper analytical push. Google’s new Gemini 3.5 Flash version should fix this blind spot, though it’s unclear whether it’s for Low or Medium effort variants.

As Google usually does as a gesture of goodwill, the company has once again completely reset Gemini rate limits for all Antigravity users. Whether you are on a free tier or a paid plan, your counters are back to zero, so you can put the updated model through its paces immediately.

It is worth reiterating that these “effort-level” variants (Low, Medium, and High) are specific to Antigravity. You shouldn’t expect to see a “Gemini 3.5 Flash Low” toggle inside your consumer-facing Gemini app anytime soon.

We will have to see if this iteration finally strikes the right balance between saving your quota and actually getting the job done.

Meanwhile, users are requesting on X that Google integrate a weekly usage bar so that they can keep track of their usage, as they can’t currently see how much weekly quota they have left and when it resets.

Mohan has taken note of this feedback, so hopefully, there’s good news coming on this end, too, in the future.