Google Unveils Gemini 4 Carbon and Genie 4 AI Models

Google announced two new large-language models this week. The model reportedly outperformed Google’s earlier Gemini 4 Argon and rival offerings such as OpenAI’s GPT-6 Astra on internal benchmarking tests that assess overall capability.

CNET

Genie 4 is presented as a “world model,” a type of AI that simulates real and imagined physical environments rather than merely predicting the next word. Google says the model is grounded in the physics of a digital universe and adds a new audio capability, allowing developers to embed background noise and dialogue in AI-generated worlds. The company cites potential uses in robotics and video-game design.

CNET

Neither Gemini 4 Carbon nor Genie 4 is currently available to the general public. Access to Gemini 4 Carbon is limited to a select group of cybersecurity experts, and Google declined to provide a timeline for broader release, noting that prior launches have opened to wider audiences within one to two weeks. Google is actively engaging with the United States government’s AI review process, though no specific milestones were disclosed. A waitlist for Genie 4 is open to developers.

CNET

The Gemini 4 Carbon launch follows Google’s previous frontier model, Gemini 4 Argon, which arrived after an 11-month gap—the longest interval for the company in an industry where new models appear weekly. Google’s announcements come amid heightened public concern about AI safety and backlash against large AI infrastructure projects such as data-center expansions. While Google remains a major AI player, analysts note that the firm has struggled to keep pace with rivals in the rapidly evolving “AI race.”

CNET

The article mentions Fei-Fei Li and Yann LeCun, two prominent AI researchers who are building world-model technologies at their respective start-ups, underscoring broader industry interest in the capabilities showcased by Genie 4.

CNET