Google puts Gemini 4 Carbon on top AI benchmark, but keeps it locked down
Google has launched Gemini 4 Carbon, its latest AI model, and says it delivers "significant gains" in coding, agentic tasks, and cybersecurity. The company claims Carbon now leads the Artificial Analysis Intelligence Index (v5.0) benchmark, ahead of Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra.
Why it matters
Google’s announcement puts Gemini 4 Carbon at the top of the current AI benchmark race, at least by the company’s own reporting. The Verge AI reports that Google is positioning Carbon as a major step forward for cybersecurity and agentic automation. The model’s performance on the Artificial Analysis Intelligence Index (v5.0) is being used as proof of its technical edge over Claude Opus 5.5 and GPT-6 Astra.
But there is a catch. According to Ars Technica AI, Gemini 4 Carbon is not available for public or business use. The model is described as "essentially a slightly better version of Argon," but Google has not opened access to customers, developers, or enterprise partners. This means the claimed improvements are not something that companies can test or integrate into their workflows yet.
How Gemini 4 Carbon compares to rivals
On paper, Gemini 4 Carbon leads the Artificial Analysis Intelligence Index (v5.0), with Google highlighting its "significant gains" in coding, agentic tasks, and cybersecurity. The company’s announcement places Claude Opus 5.5 and GPT-6 Astra in second and third place on the same benchmark. However, the details behind these scores have not been released, and no independent third-party evaluations are available yet.
Ars Technica AI notes that Gemini 4 Carbon is "essentially a slightly better version of Argon." This suggests incremental improvements rather than a major leap, at least from the perspective of outside observers. Google’s focus on cybersecurity and coding tasks may be aimed at enterprise buyers, but without access, these claims remain theoretical.
This is not the first time a leading AI company has announced a new model before making it available. Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra are both accessible to some customers, while Google’s previous models, such as Genie 4, have targeted specific commercial use cases. For a look at how Google has positioned earlier models for business, see our coverage of Genie 4.
What this means if you run a business
For business owners and operators, Gemini 4 Carbon’s launch is mostly a signal of where Google is investing its AI efforts, not a new tool to deploy. The company’s claims about cybersecurity and coding improvements may eventually matter for regulated industries or companies with complex automation needs, but as of now, there is no way to buy, test, or integrate Carbon into your stack.
The lack of access also means that any promises about better security or automation remain unproven for real-world business use. If you are already using Gemini Argon or other Google models, you will not see an upgrade to Carbon until Google changes its release policy. For now, Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra remain the top models that are actually available to enterprise customers.
Companies looking for practical AI improvements should watch for when Google opens Carbon to business or developer access. Until then, the launch is more about Google’s positioning in the AI benchmark race than about immediate business impact.
Sources
Keep up
Get the next one before everyone else is talking about it.
