Google releases Gemini 4 Carbon to everyone and unveils /swarm mode — a team of agents from a single prompt
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Google's most powerful model is now available to anyone at the price of last year's flagship, and agents have learned to work as a crew: tasks once broken into manual steps are now set with a single prompt.
Two weeks after limited access for "cyberdefenders" under the Fairwind program, Google has released Gemini 4 Carbon to general availability: via the API in AI Studio and Gemini for business for all paying developers and companies, and to all Google AI Ultra subscribers — in the Gemini app, AI Mode in Search and the Antigravity environment. The company calls Carbon its strongest model in cybersecurity and in agentic tasks that stretch over many steps: complex multi-stage programming, debugging, open-ended scientific research. The previous model, Argon, remains the recommended choice for most development and office-work tasks, while Carbon spends more tokens on deeper analysis and checks. According to Google's own figures, the model ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at more than half the per-task cost of the next-best model, and second on the new multimodal MKA benchmark (with Artificial Analysis) at more than six times lower average cost per task. Antigravity introduces a preview of /swarm: a single prompt turns into a coordinated group of agents that can be tracked in real time; Carbon takes the architecture-design, coordination and audit roles, while faster models work in parallel isolated spaces and check every change. Starting price is the same as Argon: $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 from November 30. All benchmarks and examples (assembling a 3D Eiffel Tower exhibition, a game about the Google campus) are Google's claims.