Google announced Gemini 4 Carbon on Wednesday, its latest frontier AI model and the second released in October after Gemini 4 Argon reached Pro and Ultra subscribers on October 13.
The new model tops the Artificial Analysis Intelligence Index version 5.0 with a score of 59, edging out Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra, according to the independent benchmark published on October 28.
Google DeepMind senior vice president Koray Kavukcuoglu called Carbon the company's strongest model yet in a blog post. The company said Carbon delivers a "significant leap" in cybersecurity, holding the top spot on SEC-bench Pro, which measures real-world bug discovery, and on ExploitGym, which tests the ability to turn vulnerabilities into exploits.
Carbon also surpasses GPT-6 Astra on the SEC-bench Pro cybersecurity test, The Verge reported. Google said internal teams have seen the model run for days to persist through dead ends on complex, multi-step engineering tasks.
The model is initially available only to a select group of cybersecurity defense companies through Google's Fairwind Program. Google said it will gradually expand access to more trusted partners, then to developers, businesses and consumers, though no public launch date was given.
When Carbon reaches the API, it will carry the same introductory pricing as Argon: two dollars per million input tokens and ten dollars per million output tokens, with cached input tokens at a 95 percent discount, according to The Verge.
Alongside Carbon, Google also debuted Genie 4, a world model that can simulate physical environments and now adds audio generation for background noise and dialogue. Genie 4 is accessible via a waitlist, CNET reported.
Google said Carbon cuts unsupported claims by half compared to Argon and blocked every attempt on Gray Swan's prompt injection benchmark, improving on Argon's 0.7 percent attack success rate. High-severity misalignment alerts during internal coding work dropped nearly 80 percent.
The rapid succession of Argon and Carbon reflects an accelerated pace among frontier AI labs. Anthropic released Claude Opus 5.5 on September 22, OpenAI shipped GPT-6 Astra earlier in September, and Google announced Argon on September 30 before its October 13 subscriber release, Mashable noted.