If OpenAI has adopted the solar system as its naming convention for its models, it seems Google has decided to go with the periodic table. Two weeks after the release of Gemini 4 Argon, the tech giant has just unveiled its successor, Gemini 4 Carbon, a new cutting-edge model developed with a special focus on cybersecurity and the ability to tackle much more complex, long-term engineering problems. The model's debut has been stellar, as it has achieved the top spot in the Intelligence Index of the independent analysis and comparison platform Artificial Intelligence, thanks to a score of 59, the highest recorded to date and one point higher than Claude Opus 5.5, which previously held the top spot. Improved performance for particularly complex tasks Google states that it has relied on user feedback to develop Gemini 4 Carbon, implementing performance improvements across various areas (coding, agents, automated and scientific tasks, full-stack web development, and multimodal understanding). Gemini 4 Carbon has been designed to run automated and sustained workflows, and is capable of tackling highly complex tasks that require multi-step, long-term work. “Internally, our teams have seen Carbon run for days to overcome obstacles and tackle complex, multi-stage engineering tasks, such as root cause identification and correction of subtle errors in large codebases and low-level systems,” the company explains. Cutting-Edge Cybersecurity But if anything truly defines this model, it's its cutting-edge cybersecurity performance. Its advanced capabilities allow it to cover the entire cybersecurity lifecycle, from vulnerability detection and risk validation to root cause remediation. Furthermore, this AI also excels at generating secure code, achieving a success rate twice that of its predecessor, Argon. “In internal testing, Carbon drives a significant improvement in generating secure code, and in live coding environments with our own teams, it shows reduced error and forgery rates.” Gemini 4 Carbon also demonstrates greater honesty than other company models. As Google explains: “Carbon reduces unsubstantiated claims by roughly half compared to Argon. It includes methodological warnings directly in the final product or holds blocked actions with a clear list of what is still required, rather than fabricating information or circumventing missing evidence.” In addition, the model is also more collaborative and reacts better to ambiguity. “Instead of guessing or forcing a path, Carbon retrieves relevant sources and shows discrepancies to the user before taking action; as a result, users should see a much better collaboration experience with Carbon.” Availability of Gemini 4 Carbon For now, Gemini 4 Carbon is not available to the general public. Google has announced that, starting this week, a select group of cybersecurity experts will be able to test the model, and access will subsequently be expanded to trusted partners through the Fairwind program. This will allow them to gather feedback and propose improvements, as well as refine its security. An official launch date for Gemini 4 Carbon has not yet been specified, so regular users, developers, and businesses will have to wait. Google has revealed that, in the initial launch phase, Gemini 4 Carbon will be priced at $2 per million entry tokens and $10 per million exit tokens, with entry tokens cached at a 95% discount off the entry token price. However, Google indicates that "once the introductory period ends, a price of $4 per million entry tokens and $20 per million exit tokens will apply."