Google DeepMind released Gemini 4 Carbon just two weeks after Argon. The new model topped the Artificial Analysis leaderboard with a score of 59, beating both Claude Opus 5.5 and GPT-6 Astra.
Carbon's biggest gains are in cybersecurity. By Google's internal tests, it writes secure code twice as reliably as Argon, and it scores 88.2% on SEC-bench Pro, the benchmark where models hunt for vulnerabilities.
That's why the model will first roll out only to vetted cybersecurity specialists. Developers and regular users will get access later — Google hasn't given a date. Pricing for those who do have access is the same as Argon: $2 per million input tokens and $10 per million output tokens.
Skeptics, though, see Carbon as just a slightly improved Argon.