Google Unveils Gemini 4 Carbon and Genie 4 AI Models

Google has announced Gemini 4 Carbon, its most advanced AI model to date, just two weeks after its previous release. The company also debuted Genie 4, its latest world model, this week. Neither is available to the general public yet.

Gemini 4 Carbon aims at cybersecurity

According to Google, Gemini 4 Carbon outperformed both its own Gemini 4 Argon and competitors such as OpenAI's GPT-6 Astra on benchmarks designed to measure overall model capability. Yet the company is positioning it for cybersecurity specialists, citing its strength at detecting and patching vulnerabilities. That skill has already proved valuable: last week's PHP vulnerabilities, which affected a majority of the world's websites, showed how serious such flaws can be.

For now, access is limited to select cybersecurity experts. Google declined to say when availability would widen, though previous releases suggest it could happen within one to two weeks. The company says it is actively engaging with the US government's review process.

Genie 4 brings sound to AI worlds

Genie 4 is a world model, meaning it simulates real and imagined physical worlds grounded in the physics of a digital universe, rather than predicting the next word like a chatbot. Such systems are useful for robotics and video game design. The new version adds audio, letting developers include background noise and dialogue in their generated worlds. A waitlist is open now.

World models are drawing wider attention, with AI figures Fei-Fei Li and Yann LeCun both building them at their respective start-ups.

Why it matters

Google ended 2025 as a major AI player but had not kept pace with rivals; Gemini 4 Argon was its first new frontier model in 11 months. The quick follow-up with Carbon suggests a faster cadence. It arrives as public sentiment shifts, with growing concern about job losses, misinformation and deepfakes, and backlash against AI infrastructure such as data centers.