Google's new AI, Gemini 4 Carbon, is built to find and fix security flaws

Google has announced Gemini 4 Carbon, a new AI model it calls its strongest yet, with a focus on cybersecurity. It is built to find weak spots in software, check that they are real, and fix them. The public cannot use it yet. This week, access begins with a small group of cyber defenders, meaning people who protect systems from attacks.

What it does

Google says Carbon is built for long, hard engineering jobs. Its own teams have watched it work for days, pushing past dead ends to track down and patch subtle bugs in large codebases and low-level systems. Google also says it writes more secure code and makes things up less often in live coding work with its own teams.

On its own tests, Google says Carbon more than doubled Gemini 4 Argon's success rate at writing secure code. Argon launched two weeks ago. Google also says the Chrome team found Carbon produced high quality fixes more than 10 times better than another lab's large AI model.

The scores

• Ranked #1 on Artificial Analysis' Intelligence Index, version 5.0, with a score of 59.
• 88.2% on SEC-bench Pro, which tests finding vulnerabilities.
• 54.8% on ExploitGym, which tests exploiting them.
• 71.0% on CWE-bench, which tests patching them.

Google says these outside tests were run without its safety guardrails. Most of the other claims come from Google's own internal testing, so independent results will matter. Carbon also placed #4 on WebDev Arena, a ranking for building websites.

What it costs

AI services charge by tokens, which are small chunks of text the model reads or writes. Carbon starts at $2 per million input tokens and $10 per million output tokens, the same as Argon. Reused input costs 95% less. After the introductory period, the price rises to $4 and $20. The material does not say how long that period lasts.

Who it affects

If your business runs on software, tools like this could help close security holes faster. But the same skills that find flaws could also be misused. Google says it is working with the U.S. government's voluntary process for testing models before release, and is still adjusting its guardrails.

What happens next

Google will widen access slowly to more trusted partners through its Fairwind Program. It says it wants to offer Carbon to developers, businesses and consumers as soon as possible, but gives no date. For now, most businesses will hear about Carbon through the security tools and vendors they already use.