Through a post published on the blog The KeywordGoogle has announced the new model 2.5 Flash of Gemini, the first artificial intelligence model with the “adjustable” reasonable of the company.
This aspect is especially useful for developers, who can activate, deactivate or decide the “budget” to be assigned to reasoning. The “consumer” model 2.5 Flash (Experimental) It is already available for everyone within the app (web, for Android and for iOS) of the assistant as a preview version.
Google has announced the 2.5 Flash model of Gemini
Yesterday evening, Google announced the new model 2.5 Flash of Gemini, made on the “solid” base of the model 2.0 Flash But with all the peculiarities of the 2.5 family models which, as we have seen during the announcement at the end of March, is made up of “thinking and capable of reasoning” IA models.
According to what is released by the Mountain View giant, this new model offers slightly lower performance compared to the model 2.5 Pro But in any case better than those of competing models (especially in complex tasks that require more reasoning phases).
The Gemini app welcomes “2.5 Flash (Experimental)”
Google has already started the release of the new model 2.5 Flash (Experimental) in the Gemini app (both web and for Android and iOS devices). Since all the models of generation 2.5 enjoy a “reasoning budget”, this new model (still in the experimental phase) replaces the previous one 2.0 Flash with Thinking (Experimental).
At the time of drafting this article, therefore, the models available for everyone in the Gemini app are 2.0 Flash, 2.5 Flash (Experimental), 2.5 Pro (Experimental) in limited version e Deep Research.
For Advanced subscribers, who are (gradually) receiving Veo 2 For the generation of videos, the latter is replaced by Deep Research With 2.5 Pro. In addition, the model 2.5 Pro (Experimental) It can be used without limits.
The model is already available for developers
The model 2.5 Flash It is already available for developers through Gemini bees in Google to study and vertex ai (as well as within the assistant app). The Mountain View giant invites developers to test the new parameter Thinking_budget which allows them to “check the reasoning” of the assistant.
If you want to maintain minimum costs and latency, still improving performance compared to Flash 2.0, set the thought budget at 0. You can also choose to set up a specific token budget for the thought phase using a parameter in the API or the cursor in Google to study and in vertex AI. The budget can vary from 0 to 24576 token for flash 2.5.
COme download or update the app of Google’s assistant IA
Gemini It is officially available in Italy both as a web app (on the site https://gemini.google.com/app) and as “app” for Android devices (it is always part of the Google Apps, on a par with Google Assistant), with the app on the Google Play Store app which can be reached via the badge below.
Net of the connection It is good to verify that the most recent version of Google app Which, as mentioned, is the true “container” of the assistant based on the artificial intelligence of the Mountain View giant: to do so, just make a tap on the underlying badge and, again, on “Update” in the event that the presence of an update is reported.
