Google Launches Gemini 4 Carbon with Agent Swarm Capability

Google has opened public access to Gemini 4 Carbon, its newest frontier model, after a brief limited release to cybersecurity partners. The company positions Carbon as its most capable system to date, targeting workloads that demand deep reasoning and multi‑step execution.

Alongside the model, Google previewed a “swarm” feature on its Antigravity agent platform. The feature lets a single prompt spawn a coordinated team of agents that can collaborate on long‑horizon tasks such as large‑scale code refactoring or scientific hypothesis testing.

Pricing is aggressive: an introductory $2 per million input tokens and $10 per million output tokens, roughly one‑fifth the cost of OpenAI’s GPT‑6 Astra and a quarter of Anthropic’s Fable 5.2. After the promotional period ends in late November, rates will double but remain well below rival frontier models.

Why it matters for GPU / AI infrastructure

Carbon’s token‑efficient architecture and low per‑token price reduce the compute budget for high‑throughput inference, making large‑scale agent swarms more viable on GPU clusters. For providers like AiGpu, the model’s demand profile signals growing need for flexible, high‑memory GPU instances that can sustain concurrent agent workloads without prohibitive cost.