Google releases Gemini 4 Carbon with agent ‘swarm’ feature

Google is rolling out Gemini 4 Carbon, its latest AI model, to the general public on November 10, 2026, two weeks after it released the model to a select group of cybersecurity defenders. Google claims Carbon is its strongest model yet and says it is capable of competing with the very best systems from Anthropic and OpenAI. The Verge's report carries Google's account of that public rollout, the comparison with Argon, and the swarm preview on Antigravity.

This piece covers the work Google assigns to Carbon, the work that stays with Argon, the swarm preview that turns one prompt into a coordinated team of agents, and the introductory token prices against GPT-6 Astra and Fable 5.2. It is for paying developers, enterprises, and Google AI Ultra subscribers who run complex multi-step coding, open-ended scientific research, or long-horizon agent work. Lower-tier subscribers and free users have no stated opening date in this release.

01Public rollout after defender access

Google released Gemini 4 Carbon to a select group of cybersecurity defenders, then opened the same model to the general public two weeks later, on November 10, 2026. Google claims Carbon is its "strongest model" yet. That public step puts the model in front of all paying developers and enterprises, and in front of Google AI Ultra subscribers, after the defender-only window.

Google says Gemini 4 Carbon is capable of competing with the very best systems from Anthropic and OpenAI. The claim is part of the November 10, 2026 public rollout, two weeks after a select group of cybersecurity defenders received the model. Google named paying developers, enterprises, and Google AI Ultra subscribers for that wider release.

02Argon for everyday engineering work

Google says Gemini 4 Carbon "pushes the frontier even further" than Argon, the company's other frontier model. Google launched Argon only last month. This Carbon release comes hot on the heels of that Argon launch, and Google still presents Argon as a current frontier model beside Carbon.

The company said Argon will likely remain "the ideal choice for the majority of software engineering and knowledge work." Carbon is "built for tasks that test the limits of a model's intelligence." Google names complex, multi-step coding and open-ended scientific research as the tasks Carbon is built to take.

03Coordinated agents from one prompt

To highlight Carbon's utility for long-horizon work involving AI agents, Google said it is previewing a swarm feature on Antigravity. Google calls Antigravity its agentic development platform. The swarm preview is the piece of this rollout that Google is using to show Carbon on agent work that runs for a long horizon.

The swarm feature will let users turn a single prompt "into a coordinated team of agents," and Google is previewing that handoff on Antigravity, its agentic development platform. The input Google describes is a single prompt. The result it describes is a coordinated team of agents for long-horizon work involving AI agents.

04API access for paying customers

Carbon is rolling out today via the Gemini API in Google AI Studio and in Gemini for business, for all paying developers and enterprises. Google said that in a blog post. For this November 10, 2026 release, those API surfaces are how paying developers and enterprises call Carbon.

The model will also be available to all Google AI Ultra subscribers. For them, Carbon is in the Gemini app, in the AI Mode in Search, and on Antigravity. Antigravity carries the swarm preview as well, so that platform is where Ultra subscribers meet both the model and the coordinated-agent preview.

It is not clear when, or if, Carbon will be available to lower-tier subscribers or free users. The access Google stated for this rollout covers all paying developers and enterprises on the Gemini API, and all Google AI Ultra subscribers in the app, in Search, and on Antigravity.

05Token prices and benchmark standing

As with the launch of Argon, Google is pricing Carbon very competitively. Carbon launches at an introductory price of $2 per million input tokens and $10 per million output tokens. That rate is around a fifth the cost of OpenAI's GPT-6 Astra and a quarter the cost of Anthropic's Fable 5.2, which launched yesterday, the day before this rollout.

Those costs double at the end of November, when the introductory period of lower prices finishes. Even after the prices double, Carbon will still be significantly cheaper than frontier rivals on a per-token basis. A lower price per token does not necessarily translate to lower costs, because models can use different numbers of tokens for the same tasks.

Benchmarks cited by Google when it first announced Carbon suggest the model is one of the most powerful on the market across most metrics, behind Fable 5.2. Companies can and do cherry-pick benchmark results to show their tools in the best possible light. The Verge's Robert Hart writes that the release, hot on the heels of Argon, marks a decisive return for Google to the very forefront of AI research, after months in which frontier development looked like a two-horse race between Anthropic and OpenAI.