What did Gemini 3.5 Flash actually upgrade? Its capability is still below Cursor Composer 2.5

Google Antigravity
,
1

Just tested Gemini 3.5 Flash on the same coding/problem-solving task side by side with Cursor Composer 2.5.

Gemini 3.5 Flash’s solution was shockingly bad — extremely low quality, superficial, and missing key insights. Meanwhile, Cursor Composer 2.5 delivered a clean, thoughtful, and genuinely strong approach.

And that’s not even mentioning how it compares to GPT-5.5, Claude 4.6, or Claude 4.7. The gap is massive.

So… what exactly improved in this “upgrade”? Feels more like a sidegrade or even a downgrade in practical intelligence.

Anyone else experiencing the same?

2

What improved in 3.5 is probably more reasoning. Which resulted in reduced efficiency in the name of a more agentic model. My experience is not bad. But I still think the Flash 3.0 should be available. They should have it as : Flash 3.0. 3.5, then 3.1 Pro. And separate quotas.

Im hearing good things about Composer 2.5. I will most likely switch to cursor soon enough.