Google simply launched a really full technical report of his new fashions together with its efficiency in comparison with all of his earlier ones. You possibly can test it out right here:
However summarizing, as suspected Gemini 1.5 Flash does have a lack of efficiency in benchmarks traded by his low latency and optimized prices.
This loss retains of a most of 15% much less in comparison with the opposite and it did shock me positively. As a result of the loss can solely be thought-about related beneath the context of a really advanced process and won’t make a distinction with most normal easy to intermediate duties.
Making these fashions a really accessible and highly effective possibility for making options.
It performs means higher than 1.0 PRO, plus together with his lengthy context and its uncommon skill of being able to utilizing all of it. I anticipate to see a motion of him slowly changing 1.0 PRO. Really i believe perhaps Google long run objective is perhaps do away with 1.0 Gemini household of fashions completely since 1.5 Professional performs higher than Extremely and 1.5 Flash will get actually shut evaluating the older being a a lot bigger mannequin and possibly costing mannequin (purpose why he was not launched on VertexAI for the general public to today).
We will discover this tendency whereas 1.5 Flash value beneath 128K tokens is lesser than 1.0 Professional value with the identical vary.
Now, the one unknown is Gemini Nano as a result of the unique is constituted of 1.0 batch, i ponder if they’re investing in an optimized model of it primarily based on Gemini 1.5.
