Google Unleashes New AI Model with Major Leaps in Programming and Security

On October 29th, Google officially introduced the latest iteration of its frontier AI model series: Gemini 4 Carbon. Topping the latest comprehensive AI capability rankings, this model demonstrates particular prowess in handling complex programming tasks, extended autonomous operation, and cybersecurity defense.

Core Capabilities: From Code Generation to Proactive Defense

Google's released data highlights breakthrough performance for Gemini 4 Carbon across several key security metrics. Most notably, its capability for software vulnerability discovery achieved a score of 88.2%, indicating a high-accuracy AI for identifying potential security flaws in code.

Furthermore, the model's performance in generating secure code more than doubled compared to its predecessor. In the face of growing cybersecurity threats, Gemini 4 Carbon also exhibits robust defensive characteristics: its susceptibility to prompt injection attacks has been reduced to 0%, while false positive alerts for high-severity security alignment issues have decreased by nearly 80%.

Commercial Approach: Steady Pricing, Phased Rollout

On the commercial front, Google has maintained pricing consistent with the previous model generation: $2 per million input tokens and $10 per million output tokens. For users requiring cached input tokens, Google offers a substantial 95% discount, which can help lower large-scale implementation costs for enterprises.

Currently, access to Gemini 4 Carbon is initially being granted to select cybersecurity defense professionals. Google states this phased approach allows for further refinement of the model's capabilities within critical security contexts before a broader deployment. Access will be gradually expanded based on feedback and testing outcomes.

  • Key Breakthrough: 88.2% vulnerability detection rate, secure code generation capability doubled.
  • Security Hardening: Effective defense against prompt injection attacks, 80% reduction in high-severity false positives.
  • Market Strategy: Unchanged pricing, prioritized access for security experts, with gradual expansion planned.