In September 2026 Google published a roundup of AI developments that included new models, app integrations, scientific projects and hardware. The announcement centers on Gemini 4 Argon, alongside updates across the Gemini family, tools for creators and students, and several research milestones.
Frontier models and multimodal tools
Google described Gemini 4 Argon as its newest frontier model, built with a 1-million-token context window and designed to address complex tasks such as cybersecurity defense, coding, financial research and legal drafting. Google said Argon is being released in stages and is currently rolling out to trusted cyber defenders through the Fairwind Program while the company gathers feedback on guardrails before wider availability.
Alongside Argon, Google announced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, which the company said deliver improved reasoning and coding performance at similar speed and cost to prior models. Google stated that Gemini 3.8 Flash Cyber provides autonomous vulnerability discovery and automated code patching and is available to trusted organizations via the Fairwind Program.
New voice and audio models include Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking for speech-to-speech and multi-step reasoning, and two text-to-speech variants: Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS. Google reported these models are rolling out to developers in Google AI Studio and Gemini API and appearing across Search Live, Gemini Live and Workspace products.
Apps, devices and scientific projects
Google added Connected Apps to Gemini, listing integrations with Airtable, Linear, monday.com, PandaDoc, Adobe, Picsart, Squarespace, Webflow, Peloton, SeatGeek and Experian. The company also released Lyria 3.5 for music generation, new Gemini Notebook study tools with interactive coursework and 60-second video overviews in over 80 languages, and a 12-month Google AI student plan for eligible college students.
On the research side, Google unveiled AlphaGenome Atlas, a database that predicts effects for all 9 billion possible single nucleotide variants in the human genome, and MAPL-EMIT, a deep-learning model developed with NASA JPL that Google said detects 50% more methane emissions than human experts and has identified over 23,000 previously unmapped plumes. Google also described Project Suncatcher, a test of TPUs in low-Earth orbit with Planet, and published a detailed fruit fly brain map covering over 166,000 neurons and nearly 12,000 cell types.
Product-level updates include pre-orders for Googlebook, new on-device features such as Magic Pointer and Rambler, and a Gemini app for Windows with an Alt+Space shortcut that opens a dedicated workspace for tasks including Gemini Spark. Google noted WeatherNext 3 now uses real-time satellite data, hourly refreshes and higher resolution and is integrated across Search, Gemini, Maps and Cloud.
Google also highlighted the launch of the DeepMind Institute and said its design teams were recognized by Fast Company as the 2026 Design Company of the Year. Argon remains limited to Fairwind participants for now; other Gemini 3.8 models and a range of tools are available to developers via Google AI Studio and Gemini API, and Googlebook is open for pre-order.
Original source: Google AI