Google announced Gemini 4 Carbon on October 28, reporting top results on three cybersecurity benchmarks. Access is limited to selected cyber defenders, and Google has not set a date for a wider release. Google’s announcement was also covered by Reuters and Ars Technica.

The launch adds a new contender to the race for AI tools that can find and fix software flaws. But the benchmark results come from Google, and do not establish how Carbon will perform in real-world security work.

What the scores measure

Google reports that Carbon scored:

  • 88.2% on SEC-bench Pro for finding vulnerabilities.
  • 54.8% on ExploitGym for testing whether vulnerabilities can be exploited.
  • 71.0% on CWE-bench for patching software weaknesses.

The benchmarks cover different stages of security work: finding a flaw, assessing whether it can be exploited, and fixing it. Google says it ran the evaluations without guardrails. That matters because the scores do not show how the model performs with safeguards enabled or in a company’s live systems.

Carbon has strong results in Google’s tests, but its performance outside those tests remains unverified.

Restricted access, with no public release date

Google is starting with a subset of cyber defenders through its Fairwind Program. It plans to expand access to trusted partners, while continuing to test safeguards before offering Carbon to developers, businesses and consumers. No date has been announced for broader access.

For now, Carbon is an announced model, not a generally available tool. Google says its introductory pricing will match Gemini 4 Argon’s rates:

  • $2 per million input tokens.
  • $10 per million output tokens.

Another company-reported result

Google also says Carbon generates secure code at more than twice Argon’s success rate on internal tests. Those results, like the benchmark scores, are from Google. Independent testing and wider access will show whether the gains hold up beyond the company’s evaluations.