Updated
Updated · The Verge · Oct 28
Google Launches Gemini 4 Carbon at $2 per Million Tokens as Cybersecurity Benchmarks Lead
Updated
Updated · The Verge · Oct 28

Google Launches Gemini 4 Carbon at $2 per Million Tokens as Cybersecurity Benchmarks Lead

3 articles · Updated · The Verge · Oct 28

Summary

  • Google on Wednesday released Gemini 4 Carbon, a new frontier model it says delivers significant gains in coding, agentic work and cyberdefense over Gemini 4 Argon.
  • Carbon tops the Artificial Analysis Intelligence Index v5.0 and also ranks No. 1 on SEC-bench Pro and ExploitGym, benchmarks for finding real-world bugs and turning vulnerabilities into exploits.
  • Google said the model is built for long-running engineering tasks, with internal teams seeing it persist for days, while improving resistance to prompt-injection attacks and cutting unsupported claims by 2x versus Argon.
  • The launch is initially limited to a select group of cybersecurity defense companies, with broader access planned through Google's Fairwind Program.
  • Google kept Carbon at Argon's introductory pricing—$2 per million input tokens and $10 per million output tokens—as it pushes to outpace OpenAI and Anthropic on advanced AI performance.

Insights

Can Google’s Gemini 4 Carbon really patch real-world vulnerabilities safely, or are benchmark wins hiding the hardest cybersecurity failures?
Why is Google limiting Gemini 4 Carbon to select defenders if it already claims top scores in coding, patching, and prompt-injection resistance?