Google’s September AI push spans frontier models, agents and science

In a blog post recapping its September announcements, Google said the month was headlined by Gemini 4 Argon, its latest frontier model, while also highlighting launches including Gemini 3.8 Flash, Gemini 3.8 Flash Cyber and Gemini 3.8 Live.September was a busy month for Google's artificial intelligence business, with the technology giant rolling out a new frontier model, expanding its AI agent capabilities, adding AI features across consumer products and advancing research in areas ranging from genomics to space-based computing.In a blog post recapping its September announcements, Google said the month was headlined by Gemini 4 Argon, its latest frontier model, while also highlighting launches including Gemini 3.8 Flash, Gemini 3.8 Flash Cyber and Gemini 3.8 Live.Gemini takes centre stageGoogle introduced Gemini 4 Argon with a 1-million-token output limit and advanced reasoning capabilities aimed at complex workloads such as coding, cybersecurity, financial research and legal drafting.The company is initially making Argon available to trusted cyber defenders through its Fairwind programme as it gathers feedback on safety measures before expanding access to developers, enterprises and consumers.Google also launched Gemini 3.8 Flash, which it said delivers improvements in long-horizon software engineering, multi-step problem-solving and agentic workflows while maintaining the speed and cost profile of its predecessor.A cybersecurity-focused version, Gemini 3.8 Flash Cyber, was also introduced with capabilities for autonomous vulnerability discovery and automated code patching.For voice applications, Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, allowing developers and users to interact with Gemini through more natural speech.

The models were made available through Google AI Studio and the Gemini API, as well as across products including Search Live, Gemini Live and Google Workspace.Google also expanded its text-to-speech capabilities with Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS.Gemini becomes more connectedGoogle expanded the Gemini app's ability to work with third-party applications through new Connected Apps.The integrations include productivity platforms such as Airtable, Linear, monday.com and PandaDoc, creative services including Adobe, Picsart, Squarespace and Webflow, and consumer services such as Peloton, SeatGeek and Experian.The company also brought its Lyria 3.5 music-generation model to Gemini, allowing users to create songs and audio from text prompts or images.For students, Gemini Notebook received new tools including interactive coursework, flashcards, quizzes and short video summaries in more than 80 languages.Google Labs also expanded its experimental CC product from a personal AI agent into a collaborative group agent designed to manage family calendars, shared tasks and household logistics.Google takes AI into the physical worldSeptember's announcements also extended beyond software.Google introduced Project Suncatcher, an initiative to test the performance of its machine-learning hardware in orbit.

The company plans to launch a test satellite with Planet to evaluate Google TPUs under space conditions.Google said satellites in low-Earth orbit can access up to eight times more solar power than on Earth, potentially opening new possibilities for AI infrastructure in space.The company also unveiled Googlebook, a new laptop built on Android and paired with desktop foundations from ChromeOS.

It includes on-device AI features such as Magic Pointer and Rambler smart voice dictation, along with background task processing.AI pushes deeper into scienceGoogle's September AI activity also included several research initiatives.The company launched AlphaGenome Atlas, which predicts the effects of all 9 billion possible single-nucleotide variants in the human genome.

Google said the database is designed to help researchers study rare diseases and complex genetics without requiring coding expertise.In climate research, Google and NASA's Jet Propulsion Laboratory developed MAPL-EMIT, a deep-learning model designed to identify methane emissions from space.

Google said the model detects 50% more emissions than human experts and has identified more than 23,000 previously unmapped methane plumes.Google also highlighted an AI-assisted project that mapped the brain of an adult male fruit fly, covering more than 166,000 neurons and nearly 12,000 distinct cell types.The company launched the DeepMind Institute, a platform intended to bring together researchers and thinkers from Google DeepMind, Google and the wider research community to examine the implications of artificial general intelligence and its potential risks.Weather AI gets an upgradeGoogle also introduced WeatherNext 3, its latest weather forecasting model, with real-time satellite data, hourly updates, higher resolution and improved precipitation forecasting.The model also incorporates clean-energy variables and is being integrated across Google Search, Gemini, Maps, Google Maps Platform and Google Cloud.