Google has released its latest frontier model, Gemini 4 Carbon, offering what the company calls “significant gains” in coding, agentic tasks, and cybersecurity. Announced on Wednesday, the new model claims the top spot on the Artificial Analysis Intelligence Index (v5.0), with Claude Opus 5.5 and GPT-6 Astra trailing behind.
Google Gemini 4 Carbon Tops Benchmarks in Cybersecurity
Carbon is an improvement over Argon, which Google released earlier this month, and it arrives with what the company describes as a “significant leap” in cyberdefense. According to Google, the model is better at finding vulnerabilities and generating secure code, and it currently holds the number one spot on SEC-bench Pro, a benchmark measuring an AI model’s ability to discover real-world bugs, as well as ExploitGym, which measures how a model can turn a vulnerability into an exploit.
Google says Carbon reflects the pace at which it is pushing the frontier forward. The model is built for completing longer tasks, with Google’s internal teams reporting that it can run for days, persisting through dead ends to tackle complex, multi-step engineering problems. It is also said to be better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users.
Gemini 4 Carbon Availability and Pricing
Carbon is currently only available to a select group of cybersecurity defense companies, but it will eventually expand to other partners in Google’s Fairwind Program. The model also promises more transparency and honesty, cutting unsupported claims by two times compared with Argon, according to Google.
It will launch at the same competitive introductory price as Argon, at $2 per million input tokens and $10 per million output tokens. That pricing positions Carbon squarely against rival frontier models from OpenAI and Anthropic, even as Google touts its benchmark lead in coding and security work.
With Carbon topping the Artificial Analysis Intelligence Index and leading on key cybersecurity benchmarks, Google is framing its latest model as the new one to beat — at least until the next round of the frontier race begins.
Also Read: Google Opens Gemini and Search to More Apps as AI Agents Reshape the Web and Meet Google DeepMind’s Genie 4: Real-Time World Model With Native Audio and Gemini Planning
Source: The Verge AI — read the full story there.
