New models
Google launches Gemini 4 Argon globally and announces successor Carbon
In October, Google rolled out the frontier reasoning model Gemini 4 Argon to developers, businesses and consumers, and announced Gemini 4 Carbon just two weeks later. Additionally, Genie 4 was released for interactive real-time simulations.
Google summarised in a blog post what the company launched in the AI field in October 2026. The update shows how quickly Google is releasing new frontier models: within two weeks of the global launch of Gemini 4 Argon, the announcement of successor Gemini 4 Carbon followed.
What was announced
Google released Gemini 4 Argon, a frontier reasoning model, globally via Google AI Studio, the Gemini API, Google Cloud, Search and the Gemini app [1]. The model focuses on complex coding, financial analysis and legal workflows. Additionally, Gemini 4 Argon for Robotics was released for physical reasoning capabilities.
Two weeks after Argon, Google announced Gemini 4 Carbon, described as "our strongest model yet with state-of-the-art cybersecurity and long-horizon capabilities" [1]. Carbon is now in pre-release testing with the US government and a select group of cyber defenders via the Fairwind Programme.
Google also introduced Genie 4, a world model for real-time interactive simulations with "unprecedented realism, advanced physics, lifelike human movement, and native audio" [1]. You can control scenes with natural language as they unfold. Google sees Genie 4 as an important step towards AGI, as it creates safe training environments for robots and AI agents.
Finally, Google is expanding the SynthID Detector globally in English, allowing anyone to check whether an image, video or audio clip is AI-generated. The technology also works with content from OpenAI, NVIDIA, Kakao and Apple.
Availability and pricing
Gemini 4 Argon is now available globally to developers, businesses and consumers via multiple Google platforms. Gemini 4 Carbon is in pre-release testing; a broader rollout to developers, businesses and consumers will follow after further safety testing. Prices are not mentioned in the announcement.
Genie 4 is available; demos are on the Google DeepMind blog. SynthID Detector is available globally in English. Specific access conditions or pricing models are not mentioned.
What you can do with it
With Gemini 4 Argon, you can perform complex coding, analysis and legal tasks. The robotics variant helps with physical reasoning capabilities in robotics.
Gemini 4 Carbon is specifically intended for long-term engineering and cybersecurity defence. The precise applications are not detailed, but the name and context point to advanced security tasks.
Genie 4 allows you to create and control real-time interactive worlds with natural language. Google mentions its use for safe training environments for robots and AI agents. Developers can build multimodal search functions with EmbeddingGemma 2 (740 million parameters) that work on local devices, including code, images, video and audio in one unified space.
Analysis
The rapid succession of Gemini 4 Argon and Carbon illustrates Google's high pace of innovation. For Dutch developers and organisations, this means that frontier models are becoming available ever more quickly, but also that evaluation and implementation are becoming more complex. The focus on cybersecurity with Carbon is relevant for Dutch companies that want to deploy AI for security tasks.
Genie 4 opens up new possibilities for Dutch companies in simulation, training and robotics. The combination of real-time interaction, physics and native audio makes it interesting for industrial applications, although it remains unclear how accessible the technology is for smaller parties.
The expansion of SynthID Detector to English is useful for Dutch professionals who need to verify AI-generated content, although a Dutch-language version remains absent for now.
Sources
This article was produced with AI assistance from the original sources cited and checked against our editorial policy. Editorial policy