Just two weeks after rolling out its Gemini 4 Argon, Google announced a higher-performing version named Gemini 4 Carbon, priced identically during an introductory window at $2 per million input tokens and $10 per million output tokens, as reported by theregister.com.
DeepMind Positions Carbon at the Frontier
Koray Kavukcuoglu, SVP at Google DeepMind and chief AI architect at Google, stated that Gemini 4 Carbon represents the company’s strongest model yet, specifically for long-running tasks and cybersecurity applications. The model arrives as Google works to cement its competitive standing following recent leadership transitions at its AI subsidiary.
Carbon enters a crowded field of frontier models. Alphabet CEO Sundar Pichai noted in a statement accompanying Q3 2026 earnings that the company’s first-party model APIs are currently processing nearly 32 billion tokens per minute—more than four times what the company processed a year ago. According to theregister.com, Pichai highlighted that both Argon and Carbon now lead at the frontier.
Decoding the Benchmark Rankings and Pricing Shift
On benchmark evaluations, Carbon takes the lead in SEC-bench Pro, ExploitGym, and CWE-bench. Behind Opus 5.5, GPT-6.1 Astra, and Sonnet 5.5, it secures the number four position on WebDev Arena. On Arena.ai’s leaderboard for Best Overall Agent, Carbon currently ranks second, trailing only Opus 5.5.
Model ratings biz Artificial Analysis places Google’s new model at the top of its newly formulated Artificial Analysis Intelligence Index v5, while ranking it second behind Claude Code Opus 5.5 on its Coding Agent Index. When the introductory pricing period expires, Google will lift prices to $4 per million input tokens and $20 per million output tokens—matching the pricing of Anthropic’s Claude Opus 5.5.
Targeted Deployment for Cybersecurity Defenders
Google is releasing Carbon to select cybersecurity defenders, with broader availability planned through the company’s Fairwind access program. Kavukcuoglu reported that Carbon achieves more than a 2x uplift in success rate for generating secure code compared to Argon on internal benchmarks. Furthermore, the Chrome division observed that Carbon achieved superior results by a factor exceeding 10x when compared against a large frontier model from another lab during the creation of top-tier patches.
While Google competes directly on model performance, the company maintains a massive financial lead. Emarketer principal analyst Nate Elliott told theregister.com regarding Alphabet’s Q3 earnings that Google is operating at a half-trillion-dollar run rate—nearly ten times the revenue figure of either OpenAI or Anthropic.
