Google Releases Gemini 4 Carbon, Its Most Powerful AI Model to Date
Google has released Gemini 4 Carbon, its most powerful AI model to date, exclusively for Google AI Ultra subscribers at 390 shekels per month. The release introduces advanced multi-agent capabilities as Google seeks to counter rivals OpenAI and Anthropic.
Google has announced the public release of its new artificial intelligence model, Gemini 4 Carbon, making it available to the general public after a two-week period during which it was restricted to a select group of cybersecurity experts. The company positions the new system as its most powerful model to date, featuring advanced capabilities in software coding, scientific research, data analysis, and the execution of complex tasks through autonomous AI agents.
Access to the model within the Gemini application is currently restricted to subscribers of Google AI Ultra, the company’s highest-tier subscription plan priced at 390 shekels per month, and remains unavailable to free users or those on lower-tier plans. Carbon joins Gemini 4 Argon, which was launched last month, though it does not replace it. According to Google, Argon will continue to serve as the preferred option for routine programming and standard workflow tasks, whereas Carbon is engineered for assignments requiring exceptionally advanced reasoning capabilities, such as complex software development, multi-stage debugging, and open-ended scientific research. To achieve this, the model allocates greater computing resources and time to evaluate alternative solutions, conduct in-depth analysis, and iteratively verify results. In the Gemini application interface, both models will appear side by side in the model selection menu, with Argon categorized under "Code & analysis" and Carbon designated for "Most complex work."
A notable innovation accompanying the launch is Swarm, a new capability currently in an experimental phase within the Google Antigravity development environment. Rather than operating a single AI agent to execute a task from inception to completion, the system enables the deployment of a coordinated team of agents that operate concurrently and coordinate their workload following a single prompt from the user. Each agent can be assigned a distinct segment of a project, such as software architecture planning, code authoring, testing, and troubleshooting. These sequential rollouts follow an extended period during which Google struggled to maintain momentum in the fiercely competitive artificial intelligence landscape. Following the release of Gemini 3 last year, the company stabilized its competitive standing, yet subsequently refrained from introducing a new flagship generation while competitors such as OpenAI and Anthropic continuously deployed advanced models and enhanced their capabilities. Now, within the span of less than a month, Google has released two new models from the Gemini 4 family—Argon and Carbon—in an effort to close the gap and reassert its leadership at the forefront of the industry.
