SummaryGoogle has opened its top-tier 'Gemini 4 Carbon' model to general users. Simultaneously, it offered a preview of the /swarm multi-agent orchestration feature in the Antigravity developer tool. This feature enables parallel task distribution among agents, with each agent capable of running a different model. |
Why it mattersThe general availability of Carbon signals Google's competitive move against OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5 models. This competition is particularly prominent in security and coding workloads. The /swarm feature draws attention by offering a practical multi-agent workflow where the coordinator model (Carbon) distributes tasks to cheaper worker models (Argon). This could reshape how teams structure large-scale development projects. The 50% discount on input pricing, valid until November 30, indicates that Google is actively encouraging enterprise adoption before competitors lock in contracts. |
Highlights- API pricing is set at $4/M input and $20/M output tokens; a 50% input discount applies until November 30
- /swarm allows a single prompt to spawn parallel agents; each agent can be assigned to a different model (Carbon as coordinator, Argon as worker)
- Carbon was previously held back behind the 'Fairwind Program' cybersecurity line; it is now available in Google AI Ultra, Gemini for Business, and API layers
|
Critical takeThe article is essentially a repackaged Google press release. Performance claims such as 'most powerful' and 'outperforming GPT-6 Astra and Claude Opus 5.5' are self-reported by Google without independent benchmark data. /swarm is still in preview; therefore, the reliability and token cost efficiency of the multi-agent workflow remain unverified. The 'half-price' emphasis also obscures the fact that the $20/M output token price is already at the premium end of the market. Consequently, this discount functions more as a marketing hook than a structural price change. |
This critique was written by the local AI model
Qwen3.8-27B. I chose the story; I do not review each critique before it is published.