AI Ultra subscribers can now give the latest Gemini model a spin.

Two weeks after announcing Gemini 4 Carbon for cybersecurity professionals, Google is expanding access to its latest frontier AI model to even more people. It’s now available to developers, enterprises, scientists and consumers. 

As with all new AI models, there are improvements across the board, but Gemini 4 Carbon particularly excels in areas such as knowledge work, agentic coding, science, mathematics and cybersecurity. Google says Carbon outperforms its Argon AI model and competitor models from OpenAI and Anthropic in benchmark tests across a series of areas.

The model is also said to have improved spatial reasoning and physics simulations for 3D game engines and creative coding. Previous Gemini models were already good for vibe-coding apps and games, and Carbon should be even better. 

As these frontier models become more advanced, virtually all AI companies are prioritizing the mitigation of risks of misuse and cyberattacks such as prompt injections. The model scored a 0% prompt-injection attack success rate on the Gray Swan IPI benchmark test.

The everyday user may not reap as many of the benefits from the new model, but cybersecurity experts, scientists and other professionals are more likely to make use of the improvements Google has made to its latest model. 

Gemini 4 Carbon is available for developers via its AI Studio and Android Studio, and everyday users who are subscribed to the AI Ultra plan can access the model from the Gemini App.

Source link