What do you guys think about Gemini 3 Flash?
I know that it’s glitchy and sometimes it’s doing things on its own.
But spending more time with this model, I have realised that this is a really sharp tool,
if you know how to work with it.
If you work with it in small steps, giving it specific tasks—“step by step”—then this model can work wonders.
In my opinion, this model has a very bright future.
It just needs to be calibrated in a slightly more sophisticated way.
However, I find the speed and understanding of tasks, along with solving them when broken down into steps, very satisfying and promising.
I have noticed that lately I am mostly using this model alone, and I have built a few complicated projects with it.
If you are using this model, please share your experience of working with it.
It’s fascinating to learn how other people are using it and what problems they run into, if any.
It’s working well for me. It’s taken some time to figure out to work with it effectively. I came from Firebase Studio and its in-house model, which was horrible.
Exactly.
The key phrase is “figure out.“
I really learned how to work with it.
But once you “figure it out,” it’s really working sharply.
I think that once the Antigravity team tunes it in, this model will be one of the best.
I feel like this model really works differently from other models.
I think it’s even stronger than Claude Sonnet 4.6.
This strength lies not in the task itself, but in the successful completion of the entire project.
The Antigravity team should really pinpoint this model and tune it.
This model has huge potential…
I use a multi-model technique via the Gemini CLI. If Flash can’t fix a problem, I escalate it to Pro. With multi-step processes and a “conductor” for help, Flash can complete most of my coding needs, including testing and validation.
The only problem is that even for pay-as-you-go users, the Flash model can be busy or unavailable sometimes. I hope the Gemini team releases Gemini 3.1 Flash soon.
