Gemini 1.5 Flash vs Gemini 1.5 Pro — How the model really performs?

Google just released a very complete technical report of his new models including its performance compared to all of his previous ones. You can check it out here:

But summarizing, as suspected Gemini 1.5 Flash does have a loss of performance in benchmarks traded by his low latency and optimized costs.

This loss keeps of a maximum of 15% less compared to the other and it did surprise me positively. Because the loss can only be considered relevant under the context of a very complex task and will not make a difference with most usual simple to intermediate tasks.

Making these models a very accessible and powerful option for making solutions.

It performs way better than 1.0 PRO, plus including his long context and its unusual ability of being capable of using all of it. I expect to see a movement of him slowly replacing 1.0 PRO. Actually i think maybe Google long term goal might be get rid of 1.0 Gemini family of models entirely since 1.5 Pro performs better than Ultra and 1.5 Flash gets really close comparing the older being a much larger model and probably costing model (reason why he was not released on VertexAI for the public to this day).

We can notice this tendency while 1.5 Flash price under 128K tokens is lesser than 1.0 Pro price with the same range.

Now, the only unknown is Gemini Nano because the original is made from 1.0 batch, i wonder if they are investing in an optimized version of it based on Gemini 1.5.