My experience with complex reasoning is that Gemini 2.5 Pro hallucinates way too much and it's far below gpt 5 thinking. And for some reason it seems that it's gotten worse over time.