Abstract:
On November 10, Google announced that its latest artificial intelligence model Gemini 4 Carbon will be fully open to the public. Just two weeks ago, the model was only available to a small group of "cybersecurity defenders." Google claims that Carbon is the “strongest model” to date, capable of competing with the best systems from Anthropic and OpenAI.
Google says Gemini 4 Carbon goes a step further than the company's other cutting-edge model, the Argon, which was only released last month, by "pushing the boundaries of model capabilities further." Google also stated that Argon will likely remain "ideal for most software engineering and knowledge-based work," while Carbon is "built for tasks that test the upper limits of a model's intelligence," such as complex multi-step programming and open-ended scientific research.
To highlight Carbon's usefulness in long-term tasks involving AI agents, Google also announced the preview of a "swarm" feature on its agent development platform Antigravity. This feature allows users to "convert a single prompt word into a team of coordinated agents."
Google introduced in a blog post that Carbon will be available in Google AI Studio and enterprise-oriented Gemini for business from the same day through the Gemini API, and can be used by all paying developers and enterprises. The model will also be available to all Google AI Ultra subscribers in Gemini apps, AI patterns in search, and the Antigravity platform. It’s unclear when, or even if, Carbon will be available to lower-tier subscription or free users.
As with the launch of Argon, Google has set a very competitive price for Carbon. The entry price for the model when it goes online is $2 per million input tokens and $10 per million output tokens, which is roughly equivalent to one-fifth of the price of OpenAI’s GPT-6 Astra and one-quarter of the price of the Fable 5.2 model just released by Anthropic yesterday. Even after the low-price promotion period ends at the end of November and the above-mentioned price doubles, Carbon will still be much cheaper than its competitors’ cutting-edge models in terms of tokens. However, this does not necessarily mean that the actual cost of use is lower, because the number of tokens consumed by different models to complete the same task may be different.
Benchmarks cited by Google when it first announced Carbon showed that the model was one of the most powerful on the market by most metrics, second only to Fable 5.2. It should be pointed out that companies do and often choose the test results that are most beneficial to them to display their products.
Carbon, which follows Argon closely, marks Google’s decisive return to the forefront of AI research. In the months leading up to this, the development of cutting-edge models had looked like a two-horse race between Anthropic and OpenAI.
Comments