Google launches Gemini 4 Carbon with multi-agent swarm feature
Google's newest frontier model aims to compete with OpenAI and Anthropic by handling complex, multi-step tasks like scientific research.
The short version
- Gemini 4 Carbon is Google's strongest model to date, designed for tasks that test the limits of AI intelligence.
- The 'swarm' feature on the Antigravity platform lets one prompt trigger a coordinated team of agents.
- Introductory pricing is set at $2 per million input tokens and $10 per million output tokens.
- Carbon is currently available to paying developers, enterprises, and Google AI Ultra subscribers.
Why it matters
This release signals Google's intent to compete directly with the top models from OpenAI and Anthropic on high-end research tasks. For users, it introduces a way to automate complex workflows by delegating work to multiple AI agents simultaneously.
A new frontier for intelligence
Google has officially released Gemini 4 Carbon to the general public. This follows a two-week period where the model was exclusively available to a specific group of cybersecurity defenders. The company describes Carbon as its strongest model yet, positioning it to compete with the leading systems from Anthropic and OpenAI.
While Google recently launched another frontier model called Argon, Carbon is intended for a different purpose. Argon is expected to remain the primary choice for standard software engineering and general knowledge work. Carbon, however, is built specifically for tasks that push the boundaries of what an AI can do, such as open-ended scientific research.
The model is designed to handle complex, multi-step coding projects that require deeper reasoning. By focusing on these high-difficulty tasks, Google aims to establish Carbon as a leader in the most demanding areas of artificial intelligence. It marks a significant move for the company as it seeks to reclaim its position at the forefront of the industry.
The power of agent swarms
One of the most notable additions with this release is a feature called 'swarm.' This feature is being previewed on Antigravity, which is Google's platform for agentic development. An agent in this context is an AI system designed to perform specific tasks autonomously or as part of a larger workflow.
The swarm feature allows a user to take a single prompt and turn it into a coordinated team of agents. Instead of one AI trying to complete a massive project alone, the system can break down the work and assign different parts to various agents. This approach is intended to help users manage long-horizon work that involves many moving parts.
By using a coordinated team, the system can theoretically handle more complex instructions more efficiently. This move toward agentic workflows represents a shift from simple chat interfaces to systems that can execute multi-step plans. It allows for a more modular way of interacting with AI to solve large-scale problems.
Pricing and availability
Google is making Carbon available through the Gemini API in Google AI Studio and for Gemini for business. It is also being integrated into the Gemini app, the AI Mode in Search, and the Antigravity platform. These services are currently accessible to all paying developers, enterprises, and Google AI Ultra subscribers.
The company is using a competitive pricing strategy to attract users. Carbon is launching with an introductory price of $2 per million input tokens and $10 per million output tokens. These rates are significantly lower than those of its main competitors, such as OpenAI's GPT-6 Astra and Anthropic's Fable 5.2.
While these introductory prices will double at the end of November, Carbon will still remain cheaper than its rivals on a per-token basis. However, Google notes that lower per-token costs do not always mean lower total costs. This is because different models may use a different number of tokens to complete the exact same task.
Benchmarks and competition
When Google first announced the model, it cited benchmarks suggesting that Carbon is one of the most powerful models on the market. According to these metrics, it ranks behind Anthropic's Fable 5.2 across most categories. These benchmarks are used to show how the model performs against other high-end systems in various tests.
It is important to note that companies often cherry-pick specific benchmark results to highlight the strengths of their tools. While Carbon shows high performance, the actual experience can vary depending on the specific use case. The release is a major step for Google, which had seen the frontier of AI development become a two-horse race between Anthropic and OpenAI for several months.
The arrival of Carbon signals a decisive return for Google in the high-end AI research space. By offering a model that can handle complex scientific research and multi-step coding, Google is targeting the most sophisticated users. The company is positioning itself as a primary provider for the next generation of AI-driven research and development.
What to watch
- Whether Carbon will be made available to free users or lower-tier subscribers.
- How the 'swarm' feature performs on real-world scientific research projects.
- How the pricing will affect the total cost of complex tasks compared to OpenAI and Anthropic.
Try it
If you are a paying developer, you can access Gemini 4 Carbon today via the Gemini API in Google AI Studio.
Sources and links
- Officialrolling out blog.google
- ReportingThe Verge: the article we sourced
First reported by The Verge. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.