The company's new frontier model tops Claude Opus 5.5 in several key benchmarks.

OpenAI and Anthropic's frontier models have been trading blows of late, with each claiming bragging rights in different categories. Now, Google has barged in with its latest Gemini 4 Carbon model, boasting of top benchmark scores in key areas like cybersecurity and coding. "Building on Gemini 4 Argon, which was launched two weeks ago, Carbon illustrates the pace at which we're pushing the frontier forward," the company wrote in a blog post.

The company's benchmarks show Gemini 4 Carbon hitting the highest AA Index v5.0 score (a general AI benchmark) at 59, just ahead of Claude Opus 5.5 and GPT-6 Astra. Google also claims a near sweep across categories including agentic coding, computer use, knowledge work, science and math, and others. Google especially highlighted Carbon's domination in cybersecurity, where it bested other models by fairly wide margins.

In that category, Google claims that Gemini 4 Carbon excels in finding and prioritizing vulnerabilities, as well as patching them at the root cause. The company said that it also advances frontier model safety and security by reducing shortcuts and loophole exploitation, boosting transparency and honesty (yes, there's a test for "unsupported claims"), resisting prompt injection, improving collaboration and "following through without overstepping." It's also twice as likely to generate secure code compared to Argon.

As we've seen with previous frontier models, Carbon will only be released to select cybersecurity defense companies at first, to allow for pre-release model testing.

Source link