Skip to content
VibekollenBETAVibekollen

Source

Google DeepMind

24 items in the feed. The texts are the sources' own descriptions — the content belongs to Google DeepMind.

BlogGoogle DeepMind

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

Google DeepMind introduces two new AI models for voice conversation: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Gemini 3.8 Live is optimized for scalability and cost efficiency, with the ability to process visual information in near real-time and automatically switch between 97 languages. Gemini 3.8 Live Extended Thinking is built for complex tasks and can reason and speak simultaneously, making it suitable for multi-step workflows. Both models are available to developers via Gemini API and to enterprises through Gemini Enterprise, with Extended Thinking also available to Google Workspace users.

15 Sept deepmind.google

BlogGoogle DeepMind

AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome

Google DeepMind has created AlphaGenome Atlas, a tool that maps how 9 billion possible single-point mutations in human DNA affect cell function. Instead of testing each variant in the laboratory, the system uses AI to predict the effects — a way to understand which DNA changes can cause disease and which are harmless. This makes it possible to more quickly identify genetic risk factors for diseases without needing to conduct millions of experiments.

8 Sept deepmind.google

BlogGoogle DeepMind

Introducing WeatherNext 3, our most advanced and accurate global weather AI model

Google DeepMind presents WeatherNext 3, an AI model for weather forecasting that is significantly more advanced than previous versions. The model is updated hourly with satellite data, provides forecasts in much higher resolution (5 kilometers instead of the previous 25 kilometers), and can predict local weather conditions much more accurately. It is particularly important for regions in Latin America, Africa, and the Asia-Pacific that previously could not obtain high-resolution forecasts. The model can also predict wind and solar energy production, which helps power grids plan renewable energy.

3 Sept deepmind.google

BlogGoogle DeepMind

Proactive cyber defense for governments and enterprises

Google DeepMind has launched the Fairwind Program, a limited access program for governments and businesses seeking to use advanced AI tools for cyber defense. The program provides access to Gemini 3.8 Flash Cyber alongside CodeMender, tools that can identify and automatically fix security vulnerabilities in code within minutes instead of weeks. Initial partners include government agencies, critical infrastructure operators, and major technology platforms—over 650 organizations globally are already participating. Google has also made extensive cybersecurity investments, including $36 million to support hospitals, schools, and municipal utilities in the United States.

2 Sept deepmind.google

BlogGoogle DeepMind

Introducing Gemini 3.8 Flash and 3.8 Flash Cyber

Google DeepMind introduces two new AI models in the Gemini family. Gemini 3.8 Flash is an updated coding and reasoning model that improves at solving complex programming problems and multi-step tasks, at the same price as its predecessor (0.75 dollars per million input characters). Gemini 3.8 Flash Cyber is a specialized security model that finds and fixes data security vulnerabilities in code — it finds two and a half times more correct security patches than previous competitors, according to Google's own Chrome team. Both models use the same underlying intelligence but are trained for different purposes.

2 Sept deepmind.google

BlogGoogle DeepMind

Introducing agentic video understanding with Gemini

Google DeepMind is launching a new video analysis feature for Gemini models (3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite) that reduces energy consumption by up to 88 percent and costs by up to 66 percent while increasing accuracy by up to 7 percent. Instead of analyzing video at a fixed rate, this "agentic" feature performs intelligent searches through video files — it can search for specific moments, detect anomalies, or count objects without loading the entire video. The feature is available now via Gemini API and will soon be available in the Gemini app for regular users.

1 Sept deepmind.google

BlogGoogle DeepMind

Gemini Omni 1.1 Flash lets you build with more control

Google DeepMind has introduced Gemini Omni 1.1 Flash, an updated video generation system with better controls for developers. The system can now analyze up to 10 seconds of previous context — significantly more than earlier versions that only saw the last second — enabling improved visual consistency when extending videos. New features include the ability to create transitions between keyframes, generate quick low-resolution previews for prototyping, produce polished 4K video, and use video references to maintain character appearances. All of this enables building generative video workflows and media editing programs that are faster and cheaper to iterate on.

27 Aug deepmind.google

BlogGoogle DeepMind

Piloting the world's first double-blind AI evaluations

Google DeepMind introduces the world's first "double-blind" evaluation of AI models — a method where evaluators and model creators use cryptographic technology to test models without either party seeing the other's sensitive data. The problem being solved is that AI models can be trained on the same test questions used to assess them, making results unreliable. By using Google's Confidential Computing, evaluators can test the Gemini Flash Lite model without Google seeing the test questions and without the model being able to optimize itself based on them in advance.

27 Aug deepmind.google

BlogGoogle DeepMind

Intelligent transcription with Gemini 3.5 Transcribe

Google DeepMind introduces Gemini 3.5 Transcribe, a new speech-to-text model that converts voice directly into formatted and corrected text. The model handles background noise, complex technical terminology, and removes filler words such as "um" and "ah". It comes in two versions: one for live streaming speech with less than one second latency, and one for recorded material that can identify and distinguish up to three speakers. Developers can use it via the Gemini API and Google AI Studio to build voice applications, real-time captioning, or conversation analysis. The model supports over 85 languages and can be customized for specialized vocabulary and spellings.

26 Aug deepmind.google

BlogGoogle DeepMind

Introducing Gemini 3.7 Flash

Google DeepMind presents Gemini 3.7 Flash, an updated AI model specifically designed for coding and automated agents. The model shows significant improvements compared to the previous version 3.6 Flash — including reaching 43.6 percent correct on first attempt for code generation (versus previous 34.4 percent) and better performance in web development, law, and finance. The price is halved compared to 3.6 Flash, and the model is already available via Google Gemini Spark and developer platforms.

13 Aug deepmind.google

BlogGoogle DeepMind

Putting sign language AI into users’ hands

Google DeepMind has developed SL2T, an AI model that translates sign language into text. The model is trained on over 100,000 hours of data from more than 50 sign languages and now functions in Gboard and Live Transcribe on Pixel 11 phones, starting with American Sign Language (ASL). Unlike previous attempts, SL2T understands sign language as fully independent languages with their own grammar, not just English on hands. The system tracks body movements through an on-device model and only sends coordinates to the server to protect user privacy.

12 Aug deepmind.google

BlogGoogle DeepMind

Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration

Google DeepMind has launched Gemini Robotics ER 2, an updated AI model that functions as a brain for robots. The model can see through video cameras, understand what is happening in the physical world, and plan multi-step tasks. It can also collaborate with other robots and knows when tasks are complete by analyzing video streams. According to Google, the model outperforms previous versions — it achieves 57.4 percent accuracy when tracking progress toward task completion, and 91.3 percent when identifying exact timing for switching between steps.

30 Jul deepmind.google

BlogGoogle DeepMind

Accelerating the frontiers of scientific discovery: Google’s $40M commitment to the Genesis Mission

Google commits to providing American researchers with $40 million in AI services and cloud credits for the Genesis Mission — a national initiative to double the pace of scientific discovery using AI. The contributions include access to Google's AI tools such as AlphaFold 3 (for predicting protein structures), AlphaEvolve (a coding agent), and Gemini for Government for tens of thousands of researchers at the Department of Energy's national laboratories. Researchers are already using these tools to accelerate discoveries — for example, one research group has reduced microscope calibration from 90 minutes to 13 minutes.

22 Jul deepmind.google

← All news