Google launches Gemini 4 Carbon and Genie 4 AI models: Key features, uses and availability

Google has announced two new artificial intelligence models, Gemini 4 Carbon and Genie 4, on Wednesday, 28 October, expanding its work in cybersecurity and interactive world simulation. Gemini 4 Carbon can handle complex coding needs and identify software vulnerabilities, while Genie 4 can create realistic environments with native sound and interactivity. Google has announced that the rollout will be gradual, with the Carbon version debuting in the hands of select cybersecurity experts and Genie 4 becoming available for a trial by trusted partners.

The announcements come just two weeks after Google introduced Gemini 4 Argon. According to Google’s official announcements, the new models aim to improve AI-assisted security, software development, robotics research and interactive digital experiences. However, the models are not yet widely available to the public.

Gemini 4 Carbon focuses on cybersecurity and coding

Google says Gemini 4 Carbon is the most capable of its models for complex, long-running engineering tasks. It can detect vulnerabilities, check for possible exploits and recommend code fixes. According to the company, the model is intended to solve multi-step problems, even those that cause errors in large amounts of code.

Google claimed to achieve 88.2% for vulnerability detection, 54.8% for exploitation and 71% for vulnerability patching on SEC-bench, ExploitGym and CWE-bench, respectively. Google also reported that its secure code generation and prompt-injection attacks have improved.

The company is introducing access to a group of cyber defenders as part of its Fairwind programme and is participating in the US government-organized voluntary pre-release testing programme. The development of wider access for developers, businesses and consumers will be scheduled for a later date.

Google's API pricing is $2 per million input tokens and $10 per million output tokens, which is the same as Gemini 4 Argon.

Genie 4 brings interactive worlds and native audio

Google has also unveiled Genie 4, an artificial intelligence model capable of generating explorable virtual environments that users can control in real time. Whereas traditional chatbots react to user input, it mimics the behaviour of environments, objects and characters as a scene unfolds.

Google's Genie 4 claims it can produce ambient noise, music and interactive dialogue in addition to the visuals. Elements such as weather, camera movement and objects can be changed using natural language prompts.

Google believes there are video game applications, film pre-visualization applications, educational applications and applications for robotics and simulation-based training. It is also investigating the possibility of integrating the Genie and Gemini models to enhance planning and consistency in the generated environments.

When will new Google AI models be available?

Genie 4 has entered a research preview for trusted testers. Google will continue to offer Google AI Ultra subscribers access to Google AI Ultra via the API and Google AI Studio through Project Genie and selected partners. Users interested in the project can join the waitlist via Project Genie.

Google has warned that Genie 4 is still an experimental product. Consistency can be lost in environments during longer sessions and complex physical interactions; spatial accuracy and text rendering are still not perfect. The AI-generated video and audio will be watermarked with SynthID to facilitate the detection of AI-generated content.