Google Unveils Gemini 4 Carbon, Its Strongest AI Model
Google has unveiled Gemini 4 Carbon, its strongest AI model yet, built for cybersecurity and long engineering tasks and topping benchmark charts over rivals.
Google has unveiled Gemini 4 Carbon, its newest frontier AI model. Google DeepMind describes it as the company's strongest model yet, with state-of-the-art cybersecurity skills and the ability to work on complex engineering problems for days at a time. The launch comes just two weeks after Google introduced Gemini 4 Argon, the previous flagship model. Carbon builds on Argon but pushes further on coding, security work, and long-running tasks that need many steps to finish.
What Carbon is built to do
Carbon is designed for two jobs at once: advanced software engineering and cybersecurity defense. It can find weaknesses in software, test whether those weaknesses can be exploited, and write fixes for them. Google says the model can work through a hard engineering problem for days, pushing through dead ends to track down subtle bugs across large codebases and low-level systems. It is also built to ask for human help when a task is unclear, instead of guessing or forcing a path forward.
Benchmark results
On external cybersecurity tests, Carbon scored 88.2 percent on SEC-bench Pro for finding vulnerabilities, 54.8 percent on ExploitGym for exploitation, and 71.0 percent on CWE-bench for patching. Google notes these tests were run without guardrails, showing what the raw model can do. Carbon also ranks first on the Artificial Analysis Intelligence Index version 5.0 with a score of 59, ahead of Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57. On internal tests, Google reports that Carbon generates secure code at more than twice the success rate of Argon. Its Chrome team also reported that Carbon performed more than 10 times better than another lab's frontier model when generating high-quality patches, though Google did not name the rival model. Safety also improved. Google says unsupported claims fell by about half compared with Argon, high-severity misalignment alerts in live coding environments fell by nearly 80 percent, and the model resisted every evaluated indirect prompt-injection attack.
Limited rollout first
Most people cannot use Carbon yet. Google is first giving access to a subset of cyber defenders this week, then gradually expanding to trusted partners through its Fairwind Program. The company says it is also taking part in the US government's voluntary process for pre-release model testing. Developers, businesses, and consumers will get access later, as soon as possible, after Google gathers feedback and refines its safeguards. When Carbon does reach developers, it will launch at the same introductory price as Argon: 2 dollars per million input tokens and 10 dollars per million output tokens, with cached input tokens priced at 95 percent off.
Why it matters
Carbon shows where the AI race is heading: the contest is no longer only about general chat ability, but about which model can do the hardest real work safely. Finding and fixing security flaws is one of the most valuable jobs an AI model can do, and it is also one of the riskiest, since the same skills can be misused. By limiting the first release to vetted defenders, Google is treating that risk seriously while still putting its strongest model to work. If Carbon lives up to its benchmark numbers in real use, it could change how software teams hunt for bugs - and raise the bar for every rival lab.
Sources
- Google Blog - https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/ - Reuters - https://www.reuters.com/technology/google-launches-gemini-4-carbon-ai-model-with-cybersecurity-focus-2026-10-28/