The short version
- Google has released Gemini 4 Carbon to the public, positioning it as its most capable model for high-complexity tasks like scientific research and advanced coding.
- The launch includes a new 'swarm' capability on the Antigravity platform that allows users to coordinate multiple AI agents from a single prompt.
- Pricing is set significantly lower than competing frontier models, though introductory rates will increase at the end of November.
Google has officially rolled out its latest artificial intelligence model, Gemini 4 Carbon, to the general public. This release follows a two-week period during which the system was available exclusively to a select group of cybersecurity professionals. The company describes Carbon as its strongest model to date, asserting that it is capable of competing with leading systems developed by Anthropic and OpenAI. This move signals a strategic effort by Google to reassert its position at the forefront of AI research after months of intense competition primarily between those two rivals.
The new model is designed to push beyond the capabilities of Google’s previous frontier offering, Argon, which launched only last month. While Argon remains recommended for standard software engineering and general knowledge work, Carbon is specifically engineered for tasks that test the upper limits of artificial intelligence. These include complex, multi-step coding challenges and open-ended scientific research projects that require sustained reasoning and deep analysis over long periods.
To support these demanding use cases, Google is previewing a new feature called 'swarm' on its agentic development platform, Antigravity. This functionality allows users to transform a single prompt into a coordinated team of AI agents. By enabling multiple agents to work together, the system aims to handle long-horizon tasks that require collaboration and division of labor among different specialized models.
Access to Gemini 4 Carbon is now available through several channels. Paying developers and enterprises can utilize the model via the Gemini API in Google AI Studio and through Gemini for business services. Additionally, subscribers to Google AI Ultra will find the model integrated into the Gemini app, the AI Mode in Search, and on the Antigravity platform. However, it remains unclear when or if the model will become available to users on lower-tier subscription plans or free accounts.
Google is pricing Carbon aggressively to compete with established market leaders. The introductory rate is set at $2 per million input tokens and $10 per million output tokens. This pricing structure places Carbon at approximately one-fifth the cost of OpenAI’s GPT-6 Astra and one-quarter the cost of Anthropic’s Fable 5.2 model, which was released just yesterday. Even after the introductory period ends in late November and prices double, Carbon is expected to remain significantly cheaper on a per-token basis compared to its frontier competitors.
Despite the lower per-token costs, total expenses may vary depending on how efficiently different models process specific tasks. Models can require different numbers of tokens to complete the same job, meaning that token price alone does not guarantee overall cost savings for users. Nevertheless, the competitive pricing strategy underscores Google’s intent to make high-performance AI accessible to a broader range of developers and businesses.
Benchmarks cited by Google during the initial announcement suggest that Carbon ranks among the most powerful models currently available across most performance metrics. It trails only Anthropic’s Fable 5.2 in these comparisons. However, industry observers note that companies often select benchmark results that highlight their tools’ strengths while downplaying weaknesses. Therefore, independent verification of these performance claims remains important for assessing the model’s true capabilities.
The rapid succession of releases, with Carbon following closely on the heels of Argon, marks a decisive shift in Google’s AI strategy. For several months, the frontier development landscape appeared to be a two-horse race between Anthropic and OpenAI. By launching Carbon with advanced features and competitive pricing, Google aims to break this duopoly and reestablish itself as a primary driver of innovation in large language models and agentic systems.
As the model becomes widely available, users will likely test its effectiveness in real-world scenarios involving complex problem-solving and automated workflows. The introduction of the swarm feature could particularly impact how developers design applications that require multiple AI components to interact. Whether Carbon can sustain its competitive edge against entrenched rivals will depend on both its technical performance and the adoption rates among enterprise clients and individual subscribers.
Looking ahead, the availability of Carbon to free users remains uncertain. If Google decides to extend access beyond paying tiers, it could further democratize high-end AI capabilities. For now, the focus is on developers and businesses who can leverage the model’s advanced reasoning abilities for specialized tasks. The coming weeks will provide early insights into how well Carbon performs in production environments compared to its predecessors and competitors.
Sources behind this briefing
Go to the original reporting
- The Verge↗Google releases Gemini 4 Carbon with agent ‘swarm’ feature
