It’s a shame that some of these models are being depreciated. I was using gemini-2.5-flash-lite-preview-09-2025. for a specific task in my app because it was the only one that could accurately and consistently give me the output/response needed without hallucinations. Gemini 3 flash hallucinates 90% of the time and has cost millions of tokens in bad responses for the same simple task (thousands of logs back this up). I’ve been trying 3 flash in other tasks of the app and the only thing that it can consistently produce error free is svg. gemini 3.1 would be to cost prohibitive and in testing the past few days the latency and outputs are awful for the specific use cases.