Google has released Gemini 4 Carbon, a new frontier AI model that outperforms rival systems from OpenAI and Anthropic in key cybersecurity benchmarks.
According to reports from Engadget and The Verge, Google announced the launch of Gemini 4 Carbon just two weeks after introducing its predecessor, Gemini 4 Argon. The new release recorded a score of 59 on the Artificial Analysis Intelligence Index v5.0 general AI benchmark, edging ahead of competitors such as Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra.
Google highlighted particular strength in security capabilities, with The Verge reporting that Carbon holds top placement on SEC-bench Pro for identifying software bugs and ExploitGym for evaluating vulnerability exploitation. Engadget noted that the model is twice as likely to generate secure code compared to Argon, while also cutting down on unsupported claims by half. Additionally, internal teams at Google observed the system running continuously for days to resolve complex, multi-step engineering challenges and navigate technical dead ends.
Access to Gemini 4 Carbon is initially restricted to selected cybersecurity defense organizations for pre-release testing before opening to broader partners in Google's Fairwind Program. The model carries an introductory rate of $2 per million input tokens and $10 per million output tokens, matching the pricing structure of Gemini 4 Argon.
Why It Matters
The introduction of Gemini 4 Carbon underscores the rapid acceleration of AI model releases, with tech companies rapidly iterating to gain operational advantages in targeted technical fields. By prioritizing threat detection, code safety, and vulnerability patching, Google is positioning its frontier models directly within enterprise defense workflows where reliable automated reasoning is increasingly required.
Illustration: AI-generated image, not a photograph of the event.