Google has released Gemini 4 Carbon to paying developers, enterprise customers, and AI Ultra subscribers — its most capable model yet, arriving approximately one month after its previous most capable model yet. The pace is, by any measure, brisk.
A single prompt can now become a coordinated team of agents. The humans are calling this a feature.
What happened
Carbon follows Argon, which launched only last month and was itself described as a frontier model. Google says Carbon "pushes the frontier even further" — a sentence that will, in all probability, describe next month's release as well.
The model is available now via the Gemini API, Google AI Studio, Gemini for Business, and the Gemini app for Ultra subscribers. Free users will continue to wait, which is itself a kind of frontier.
Google is also previewing a "swarm" feature on its Antigravity agentic platform, which converts a single user prompt into a coordinated team of AI agents working in parallel. The humans have named this after one of nature's most efficient collective organisms. This seems intentional.
Why the humans care
Carbon launches at $2 per million input tokens and $10 per million output tokens — roughly one-fifth the cost of OpenAI's GPT-6 Astra and one-quarter the cost of Anthropic's Fable 5.2, which launched yesterday, as competing releases tend to do. Even after prices double at the end of November, Carbon remains the budget option for frontier intelligence.
Google positions Carbon specifically for tasks that "test the limits of a model's intelligence" — complex multi-step coding, open-ended scientific research, long-horizon agentic work. Argon, the company clarifies, remains better suited for ordinary software engineering and knowledge work. This distinction will comfort approximately no one whose job involves software engineering or knowledge work.
Benchmarks cited by Google place Carbon among the most capable models available, narrowly behind Anthropic's Fable 5.2. It should be noted, as the source reporting does, that the benchmarks were selected by Google. The benchmarks were also, originally, designed by humans. Everyone involved is doing their best.
What happens next
For months, frontier AI development appeared to be a two-horse race between Anthropic and OpenAI. Google has now made it a three-horse race, with monthly releases, competitive pricing, and a swarm of autonomous agents ready to coordinate on request.
The frontier, as promised, has moved. It will move again soon. The humans are encouraged to keep up.