What events in the field of AI took place during the week
The field of artificial intelligence (AI) is developing daily. Large companies are announcing new models and their versions, adding unique features and elements for a more improved and well-coordinated work of bots. Izvestia has compiled a selection of news in the field of new AI technologies for the week.
Google unveils new AI model Gemini 3 Flash
This model is cheaper and works faster than previous versions. Currently, it is already available to millions of users in the Gemini app and through AI Mode in the search engine.
Experts note that since the launch of Gemini 3, the company has started processing more than 1 trillion tokens per day. The new version is claimed to have retained the thinking level of the Pro model, as well as received reduced latency and is more budget-friendly to use. In addition, this model surpassed Gemini 2.5 Pro in a number of indicators.
Gemini 3 Flash scored 90.4% on the GPQA Diamond test and 33.7% on the Humanity's Last Exam without using tools. On the MMMU Pro test, the model scored 81.2%, which is comparable to the Gemini 3 Pro result. At the same time, the new version surpassed Gemini 2.5 Pro in a number of indicators — it works three times faster, on average uses 30% fewer tokens to complete tasks.
Valery Cherepennikov, former CEO of Intel in Russia, also clarified in an interview with Izvestia that in some cases the model is superior to OpenAI's ChatGPT. Against the background of improved performance, the expert called Gemini 3 a radical breakthrough.
The Gemini 3 Flash model was installed by default in the application, and all users got access for free. She is not only able to analyze videos and photos, create plans, recognize drawings and generate applications, but also handles complex questions and demonstrates results taking into account all aspects of the query.
OpenAI has released a new version of ChatGPT Images
It is developed on the basis of the flagship image generation model. Allegedly, the system is now able to make precise edits, saving details, and generate images four times faster. Along with the model, a new images interface was launched in ChatGPT.
Currently, the new version is already available to all users. The model is able to work with different types of editing, including adding, deleting, combining, mixing, and moving objects. So, during editing, important elements of the image are saved.
The Disco product for creating web applications is presented.
A new AI product for the Gemini-based Disco browser has been introduced, which is capable of converting open tabs into web applications. Initially, the feature will only be available to a limited number of people through Google Labs.
Using the model, you can create a GenTabs tool that offers interactive applications for tasks related to viewing content. Text queries also allow you to create web applications.
So, in the course of studying a specific topic, GenTabs may suggest creating an application to visualize information. At the same time, the product focuses not only on the current tab, but also among other viewed pages.
Google has added the function of fitting clothes through selfies
Google has announced an update to the AI-fitting function. Previously, this was only possible through a full-length image, but now it's enough for users to upload selfies. The system independently creates a full-length digital image to generate a fitting.
To get a more accurate picture, users can choose their clothing size. As a result, the system generates several photos. At the same time, the possibility of using full—length photos has not been removed - a choice of different models is also available for this function, depending on the type of figure.
All important news is on the Izvestia channel in the MAX messenger.
Переведено сервисом «Яндекс Переводчик»
