Google introduces Gemini 4 Argon and more AI in September

Google closed out September 2026 with the announcement of Gemini 4 Argon, an AI model built for complex tasks such as programming, financial research, legal writing and defending against cyberattacks. Its main new feature is a 1 million-token context window, allowing it to analyze large amounts of information at once.

Availability will be gradual. Argon is being tested first with trusted cybersecurity defenders through the Fairwind Program, while Google gathers data to refine its safety measures before bringing it to developers, businesses and consumers.

A model built for long, complex tasks

A context window is the amount of text, code or data a model can consider during an interaction. With 1 million tokens, Argon can work with lengthy documents, entire codebases or large volumes of information without forcing you to split them into many parts.

Google also introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The first improves reasoning, programming and multistep tasks while maintaining the speed and cost of 3.7 Flash. The second is designed to detect vulnerabilities and autonomously propose code patches, although it is currently available only to trusted organizations.

That does not mean AI will repair any system without supervision. At this stage, Google is testing how far it can go and what controls it needs before expanding access.

More voice, music and connected apps

The company expanded its voice capabilities with Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two audio models that enable more natural conversations. The second version is designed for multistep reasoning without adding as much delay to responses.

These tools are already available to developers through Google AI Studio and the Gemini API, and are also coming to services such as Search Live, Gemini Live and Google Workspace. In Gmail, Docs and Keep, you can use voice to complete tasks without constantly switching between apps.

Google also launched text-to-speech models that let you create custom voices and direct dialogue from written instructions. With Lyria 3.5, Gemini can generate songs and musical arrangements from text or an image, both in the Gemini app and in Google AI Studio, Google Flow Music and Google Vids.

Another new feature is Gemini's connected apps. You can link services such as Airtable, Linear, monday.com, PandaDoc, Adobe, Picsart, Squarespace, Webflow, Peloton, SeatGeek and Experian. The goal is to carry out tasks across multiple platforms from a single conversation instead of jumping between tabs.

Tools for studying and organizing daily life

Gemini Notebook now includes interactive flashcards, quizzes and 60-second video summaries in more than 80 languages. These features are designed to turn notes and study materials into exercises tailored to each topic.

Eligible university students can also claim 12 free months of the Google AI student plan, with higher usage limits.

In Google Labs, CC is moving beyond being only a personal agent and is beginning to be tested as an assistant for groups. It can have its own email address, coordinate family calendars, track shared tasks and organize household logistics with built-in privacy controls.

For people who work on Windows, the Gemini app can be opened with the Alt + Space shortcut. From there, you can search documents, summarize Gmail and Drive files, delegate multistep tasks to Gemini Spark and generate images or videos without leaving your computer.

AI for science, climate and computing

The announcement also includes projects that are not designed as everyday products. AlphaGenome Atlas brings together predictions about the effect of the 9 billion possible single-nucleotide variants in the human genome. The goal is to support research into rare diseases and other genetic problems without requiring programming knowledge.

In collaboration with NASA's Jet Propulsion Laboratory, Google developed MAPL-EMIT, a model that detects methane emissions from space. According to the company, it identifies 50% more emissions than human experts and has already found more than 23,000 unmapped leaks, including 24 of the 25 highest-emitting landfills. The data is available in Google Earth Engine.

Google also introduced WeatherNext 3, a weather model that incorporates real-time satellite data, hourly updates and more accurate precipitation forecasts. Its results will be integrated into Search, Gemini, Maps, Google Maps Platform and Cloud.

With Project Suncatcher, the company plans to test its tensor processing units, or TPUs, on a satellite developed with Planet. In low Earth orbit, satellites can receive up to eight times more solar energy than on Earth, although the project still needs to determine whether running AI infrastructure in space is viable.

What changes for you

Most of these developments will not reach all users immediately. What is becoming clear is the direction: Gemini wants to move from answering questions to working with your documents, apps, voice, calendar and chained tasks.

Google is also bringing AI into areas where results can be measured with concrete data: detecting methane leaks, studying mutations, improving weather forecasts or analyzing large biological maps. The next step will be to see whether these capabilities work reliably outside demonstrations and what level of supervision they need before automating important decisions.