By skyforbes I am a scientist and try out alternative variations of LLMs. To do this I take open source LLMs like Qwen2.5 and train them with some own functions. I try this out for different data sets and benchmarks.
GPT5.1 is good for that and can give me useful code when I explain carefully what I want. Claude Opus is even better, but has massive token limits even when I pay.
Gemini 3 is kind of disappointing. It does not get what I want.
QUestion here: I tried AI studio and Antigravity, both are not so helpful. Are there better ways to do coding with Gemini 3?