Google

On October 29th, according to The Information quoted sources, OpenAI plans to procure a self-studyed TPU artificial intelligence chip from Google. At the same time, OpenAI has issued recruitment positions for engineers who can lead the optimization of the frontier model on TPU. It was reported that this adaptation was still at an early stage and that the position would have a broad technical responsibility for the project. Recruitment requires both training and reasoning experience, with emphasis on optimizing the real TPU workload. OpenA..

On October 10, Google officially released a new generation of Gemini 4 “Argon” models to the public, according to Business Insider. At the same time, its in-house staff are already testing a later iterative version with more performance, which is expected to further narrow the technology gap with the competition. According to internal files and screenshots obtained by the media, Google staff have recently been testing a new version of Gemini 4 codenamed “Carbon”. The model is likely to be a new check point under the Argon structure, and has recently accessed the Valley..

On September 18, the source @lyraxana posted a tweet on platform X on September 16, in which Google is based on the DeepThink V3 model, building an experimental AI model with the internal code name Mathematica, which seeks special optimization, mainly for heavy computing and complex symbol questions, and is expected to be the best mathematical reasoning AI model for Google. API data indicates that the internal identifier of the model is "models / deepthink-mathematica-tf-raw..

On August 15, Google released a tweet on platform X yesterday (14 August) for the purpose of adding Media Watermark to Gemini in the next few days to allow users to close the visible watermarks of images, videos and music. The user can find Media Watermark in entry 3 of the penultimate menu after opening Settings in the Gemini leftbar. After hovering the mouse over this option, the interface displays On and Off2 options. Keep it on, Gemini..

On August 3rd, Google announced on July 31st the opening of Gemini Spark ' s smart body to most users around the world, a product that could, at the user ' s direction, complete the process of organizing inboxes, summing up correspondence mail, cancelling subscriptions, etc. 1AI understands that Spark is the individual AI smart body of Google. The product is now integrated with the Chrome browser, and, with the authorization of the user, the smart body can use a saved account password to replace the user with a number of complex web operations such as search flights, booking tickets, searching for the collection house, etc. I don't know

On 2 August, according to the information received, the director of the OpenAI Codex project, Thibault Sottiaux, had worked at the DeepMind Laboratories for six years and had studied projects such as AI infrastructure and machine learning engineering. Today, Teebo announced on X platform that DeepMind had made a similar product one year before ChatGPT, when it was called LMChat, and changed the internal code. Ultimately, however, the product was not published. The reason is the company's concern that the product might be related to..

On July 21st, according to The Insider of Business, Google DeepMind CEO Demes Hashabis stated at a London business conference that AI was changing the career path of the technology industry, but that STEM degrees and computer science bases would not lose value. Hashabis believes that those who master the basics of software engineering and computer science can use AI tools more efficiently. He mentioned that programming languages had evolved from machine code to C language to Python, and that there might be a further shift towards programming in English. But..

On July 1, local time, on Tuesday, Google released a new self-study of AI images and video generation model Nano Banana 2 Lite, claiming that it was faster and less expensive to use than the previous generation. Nano Banana 2 Lite can generate an image in 4 seconds, with a significantly lower delay than before, suitable for work requiring rapid and repeated modification of the programme, or a large number of mass images in a short time, with a resolution of 1K (1024x1024px) with an output image consumption of 1120 token at 0..

On May 20th, Google officially released the Gemini Omni model at today's 2026 Google I/O Conference of Developers. Demis Hassanis, Nobel Laureate and Head of Google DeepMind, stated that Gemini Omni was the most capable version of the Gemini model family to date. "Omni" means "whole power" and the model displays unprecedented fluidity and depth when dealing with a variety of mosaics such as text, image, video, audio..

ON 23 APRIL, ACCORDING TO THE INSIDER OF BUSINESS, ON WEDNESDAY, LOCAL TIME, GOOGLE REPORTED THAT 3/4 OF THE NEW CODE CURRENTLY BEING DEVELOPED WITHIN THE COMPANY WAS GENERATED BY AI AND THEN SUBMITTED TO HUMAN ENGINEERS FOR REVIEW. THIS PROPORTION HAS BEEN RISING STEADILY IN RECENT YEARS. GOOGLE REVEALED THAT AS OF OCTOBER 2024, ABOUT A QUARTER OF THE COMPANY'S CODE WAS GENERATED BY AI. BY THE AUTUMN OF LAST YEAR, THIS PERCENTAGE HAD RISEN TO 501 TP3T. GOOGLE HAS BEEN PUSHING EMPLOYEES TO USE AI FOR PROGRAMMING AND OTHER WORK. CEO PICHAY REVEALED IN HIS BLOG ON WEDNESDAY THAT GOOGLE IS TURNING TO "REAL..

On April 23rd, Google (21st of April) published an article announcing the upgrading of its autonomous research intelligence, and the launch of Deep Research and Deep Research Max, based on the Gemini 3.1 Pro model, to transform it from a complex summary engine to a basic component of an enterprise-level workflow. 1AI quotes the Boeven presentation that Deep Research smarts apply to interactive user scenes, balancing optimized speed and efficiency; and Deep Research Max fits..

On March 26, Google announced the launch of the state-of-the-art music generation model, Lyria 3 Pro, which is a significant improvement over the previous generation, mainly in terms of structure and length of production. Google indicates that Lyria 3 Pro is able to more accurately process preludes, lead songs, side songs, bridge segments, etc., and supports the generation of a full track of up to three minutes, which is closer to the end-of-life music that can be used directly. At the same time, the model does not directly mimic a particular artist's style, and even if the user enters the artist's name in a hint, the model only “as an inspiration”. Training the data community..

According to news from 20 March, according to Bloomberg, yesterday Google began to share with some participants in the consumer testing programme an early version of the Gemini for Mac application, codenamed “Janus”, in private, to collect feedback and pre-issue questions. Google explained to the tester that "this is an early version of the Gemini for Mac application to collect user feedback and currently contains only key functions for other clients, not all. It's a matter of concern that Google is testing a test for Gemini for Mac called..

On February 26, Google launched a new set of Android features, including more Gemini-based tools and improved fraud detection, at today's Samsung Galaxy S26 series launch. The new functionality that Gemini is applying will allow users to transfer multistep tasks (e.g. booking of a car or adding a shopping cart) to the system. This feature will first be available as a test version, which will run backstage when the user performs other tasks. Users can monitor Gemini's progress in real time by notifying them to keep track of what they're doing..

On February 25th, Google Labs officially joined its camp when it published a book yesterday, February 24th. The ProducerAI platform, which integrates advanced large models from Google DeepMind, such as Gemini, Lyria 3, Veo, is dedicated to providing full-link creative support to musicians, ranging from writing words, composers and mixers. Users simply enter simple commands like "Producing a lofi beat", is..

On February 24th, ISTE and ASCD, Google United Education, launched a plan to provide 6 million teachers in the United States with free Gemini AI training. 1AI refers to Google Press Releases, which cover the use of Google's own AI products, Gemini and NotebookLM, aimed at helping teachers themselves and 74 million students to use AI safely in the classroom. Courses are designed to be simple and practical and provide concrete examples. In addition, this project is expected to be launched in the next few months

On 6 February, yesterday, the Google parent company Alphabet published the fourth quarter of 2025 and the entire year's financial report, showing that the revenues and profits were fully exceeded. At the same time, Alphabet has dramatically increased its capital spending this year to between $17.5 billion and $18.5 billion, almost twice as much as last year, one of the most radical AI capital investors in the world’s technology giants. Alphabet received $113,828 million in the fourth quarter, an increase of 181 TP3T; net profit of $34,455 million, an increase of..

On January 19, Google officially released an article this week (January 15) on the official web, which revealed the name of the recent fire “Nano Banana” AI production model. In late July last year, the Google team was working on the Gemini 2.5 Flash Image model, Bug, running an evaluation and preparing for the first official online version, when the model, although technically coded, lacked a public name for the LMArena platform. As a reference, LMArena is a public..

On 19 January, according to news, with Google Gemini ' s rising human atmosphere, the advertising industry has been speculating that this application may be about to be introduced to capture the current heat and help offset the high cost of artificial intelligence infrastructure. But Dan Taylor, Vice-President of Google Global Advertising, did not think so, and in an interview with Business Insider, Taylor reiterated that “there are no plans to advertise in Gemini applications”. On the contrary, the advertising team is focusing on the ads of artificial intelligence..

On January 16th, Google released a book yesterday, January 15, which, based on the Gemma 3 architecture, launched the TranslateGema Open Translation Model series, with a total of 4B, 12B and 27B parameter sizes, supporting 55 core languages and multiple mode image translations, which are now available for download in Kaggle and Hugging Face. In terms of performance, Google teams use the WMT24++ benchmark (55 languages in high, medium and low resource languages) and MetricX..

On January 15, local time 14th, Google announced that a new feature called “personal intelligence” had been opened for testing in personal accounts. This function integrates information in applications such as Gmail, Google Photo albums, helps Gemini to understand context relationships without clear guidance and allows chat robots to cross-apply to understand user data, thus providing a more personal response. Google Lab and Gemini's application vice president Josh Woodward said Gemini was able to establish a connection between the different content, whether it be electricity..

On the 11th of January, Google Official sent a letter this week (8 January) announcing that Gmail e-mails are officially online for several AI functions, called “Bring Gmail into the Gemini Generation”. The new feature, it was described as the “AI Overview” (Note: AI Overviews), which automatically integrates the mailing group with dozens of responses into a brief summary, allowing users to leave the past through the list of messages and the complex experience of collating answers from line to line. Users need only use their natural language to send Gmail..

According to December 20, according to the technology media Phone Arena today, Google Gemini has added an AI video detector to help users judge whether the video is actually taken or generated. In conjunction with an earlier report by 1AI, Gemini has been able to test AI images in November of this year, and now it is being further expanded to video. Using Gemini to test AI videos is very simple. Users need only click Gemini App on their cell phone or access web pages on their computer, and then..

In the morning of December 18th, Google Gemini 3 Flash was officially released, with a direct response to the flagship models of OpenAI and Anthropic, officially known as 2.5 Pro three times faster, at a price of one quarter of 3 Pro, without a reversal. Starting today, users can experience Gemini 3 Flash (Fast/Thining) and Gemini 3 Pro in the Gemini product line. The baseline test results show that..