AI
Google debuts Gemini 4 Carbon AI model with coordinated agent swarms
The company's latest frontier model rolls out publicly after weeks of exclusive access, bringing competitive pricing and multi-agent coordination capabilities to developers and enterprise users.
1/2
The Verge
Google has made its newest large language model, Gemini 4 Carbon, available to the broader developer community and paying users following a beta period with select cybersecurity professionals. The company positions Carbon as its most capable model to date, designed to tackle complex challenges that push AI capabilities to their limits—from intricate, multi-layered coding problems to exploratory scientific work.
New capabilities and positioning#
Carbon arrives with a fresh emphasis on agent-based workflows. Google is introducing a "swarm" feature on Antigravity, its platform for building AI agents, allowing developers to translate a single instruction into a coordinated network of specialized agents working in parallel. This advancement reflects a broader industry shift toward autonomous AI systems that can decompose and execute intricate multi-step tasks.
While Google frames Carbon as a step beyond its predecessor Argon (which launched last month), the company notes that Argon remains the practical choice for software development and routine knowledge work. Carbon is explicitly engineered for scenarios where AI reasoning must stretch further.
Availability and pricing strategy#
The model is now accessible through the Gemini API and Google AI Studio for paying developers, via Gemini for Business for enterprises, and through the Gemini app for AI Ultra subscribers. It's also integrated into Search's AI Mode and available on Antigravity.
Pricing is aggressive: Google is offering an introductory rate of $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 that of Anthropic's newly launched Fable 5.2. Even after the introductory discount expires in late November, Carbon will undercut frontier competitors on a per-token basis, though total expenses can vary depending on how efficiently each model processes identical tasks.
Competitive landscape#
According to benchmarks Google cited at Carbon's initial announcement, it ranks among the top performers across most metrics, though Fable 5.2 edges ahead on several measures. The release marks a notable shift in AI development, with Google reasserting its position at the frontier after months during which Anthropic and OpenAI appeared to dominate cutting-edge progress.

