Gemini 3 is very good at the very first output but after 3 or 4 iterations it starts to hallucinate. And from there everything is downhill.
The outputs start to become worse and you feel the first was the best, just go with that.
What differentiates latest version of each model is how good and accurate is their first output (obviously determined by how good has been the input), after that every model hallucinated in errirly similar way.
The outputs start to become worse and you feel the first was the best, just go with that.
What differentiates latest version of each model is how good and accurate is their first output (obviously determined by how good has been the input), after that every model hallucinated in errirly similar way.