Google Unveils Gemini 4 Carbon, Its Strongest Model Yet — for Cyber Defenders First

Google DeepMind unveiled its new frontier AI model, Gemini 4 Carbon, on 28 October, just two weeks after Gemini 4 Argon. Google calls it its strongest model yet, built for cybersecurity and long-horizon work — but its first users will not be the general public. They will be the defenders who protect the software everyone else relies on.

Defenders first, everyone else later

Google said Carbon will go first to a subset of vetted cybersecurity defenders, then expand to more trusted partners through its Fairwind Program. The company is also taking part in the US government's voluntary pre-release testing and says it will not open access broadly until safeguards have been evaluated. Reuters confirmed the staged rollout. Google has given no firm date for general availability to developers, businesses or consumers.

Benchmarks, and their limits

On the independent evaluator Artificial Analysis's Intelligence Index v5.0, Carbon scores 59, ranking first — ahead of Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57, with Argon at 50 on the same version. Google also says Carbon reaches 88.2% on SEC-bench Pro for finding real software vulnerabilities and 71.0% on CWE-bench for fixing them at the root cause, generates secure code at more than twice Argon's success rate in internal tests, and makes about half as many unsupported claims as Argon. Apart from the Artificial Analysis index, most of those figures come from Google's own internal testing and have not yet been independently verified.

Pricing and the race behind it

Google set introductory API pricing at $2 per million input tokens and $10 per million output tokens, with eligible cached input tokens discounted by 95%; standard rates will later rise to $4 and $20. The launch comes as Google races OpenAI and Anthropic: two frontier models in two weeks shows the contest is shifting from simply being smarter to being able to run real engineering and security jobs for days without going wrong.