Google Opens Gemini 4 Carbon to Everyone: The Model Was First Exclusively Used by Cybersecurity Defenders for Two Weeks

Google has officially made Gemini 4 Carbon accessible to the public. Two weeks ago, this model was only provided to a small select group of cybersecurity defenders. Google calls Carbon its most powerful model ever, with performance that can rival the best systems built by Anthropic and OpenAI.

As reported by The Verge, Google positions Carbon as a higher-tier offering than the Argon model launched last month. While Argon is also a cutting-edge model, Google says it will likely remain the ideal choice for most software engineering and knowledge work. Carbon is designed for tasks that push the absolute limits of AI model intelligence. This division of roles is notable: the same company releases two state-of-the-art models at the same time, one covering regular daily work scenarios, the other pushing the upper ceiling of model capabilities.

The rollout order is also quite telling. Instead of opening access to developers or regular users first, Google handed Carbon over to cybersecurity defenders in advance. Google has not published full details on exactly what this group of testers did with the model, but this sequence clearly demonstrates that the more advanced a model is, the earlier its safety assessment and controlled deployment will be prioritized.

For ordinary users, the change is straightforward: access to the model, once restricted to a tiny group, is now open to the public at large. For developers, whether Carbon is worth switching to depends on whether their tasks truly require that higher performance ceiling. Google's description of Carbon as its "most powerful model" is an internal claim, and its real-world performance still needs to be tested against specific practical use cases. Argon remains available, and is framed as the default pick for most scenarios, which at the very least makes it clear that Carbon is not a universal one-size-fits-all replacement.