Google unveils Gemini 4 Carbon, for now only for cybersecurity specialists
Google DeepMind has announced its new frontier model, which for now is used only by a small group of cyber-defence professionals. There is no date for the general public.
On 28 October 2026 Google DeepMind announced Gemini 4 Carbon, which it describes as its most capable model to date and the successor to Gemini 4 Argon, released two weeks earlier. According to the company, it begins to become available this week to a subset of people working in cybersecurity defence and will gradually be extended to more trusted partners through the Fairwind programme.
Access for developers, businesses and the general public will come “as soon as possible”, according to Google, but the announcement gives no date. For teachers, then, this is for now news about what is to come, not a tool that can be used in the classroom. That was precisely what was pointed out in the group: for the moment it is only an announcement.
Google attributes to the model improvements in programming, multi-step tasks and web development, with its biggest advance in cybersecurity. Among the figures it offers are first place in the Artificial Analysis Intelligence Index v5.0 (59 points) and more than double the success rate of Argon in secure code generation. These are results published by the company itself, with no independent verification in what was consulted.
API pricing, once it is generally available, will be $2 per million input tokens and $10 per million output tokens during an introductory period, and $4 and $20 afterwards. The announcement makes no mention of educational uses.
Google unveils Gemini 4 Carbon, for now only for cybersecurity specialists
Google DeepMind has announced its new frontier model, which for now is used only by a small group of cyber-defence professionals. There is no date for the general public.
On 28 October 2026 Google DeepMind announced Gemini 4 Carbon, which it describes as its most capable model to date and the successor to Gemini 4 Argon, released two weeks earlier. According to the company, it begins to become available this week to a subset of people working in cybersecurity defence and will gradually be extended to more trusted partners through the Fairwind programme.
Access for developers, businesses and the general public will come “as soon as possible”, according to Google, but the announcement gives no date. For teachers, then, this is for now news about what is to come, not a tool that can be used in the classroom. That was precisely what was pointed out in the group: for the moment it is only an announcement.
Google attributes to the model improvements in programming, multi-step tasks and web development, with its biggest advance in cybersecurity. Among the figures it offers are first place in the Artificial Analysis Intelligence Index v5.0 (59 points) and more than double the success rate of Argon in secure code generation. These are results published by the company itself, with no independent verification in what was consulted.
API pricing, once it is generally available, will be $2 per million input tokens and $10 per million output tokens during an introductory period, and $4 and $20 afterwards. The announcement makes no mention of educational uses.
