Google unveils Gemini 4 Carbon, an AI model built to find security flaws

Google has launched Gemini 4 Carbon, its second frontier AI model in a month after Gemini 4 Argon, promising major gains in cybersecurity, coding and agentic tasks. Google says it beat rival models from OpenAI and Anthropic in almost every test and can persist on complex tasks for days. Access is initially limited to select cybersecurity defence firms, priced at $2 per million input tokens.