Gemini 4 Carbon reaches Agent Studio with 1M output tokens

Gemini 4 Carbon, which Google calls “our most capable model for deep work,” is rolling out on Google Cloud Platform and is already available in Agent Studio, according to a post on Threads published on November 10. Google set Carbon at Argon’s introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% below the input rate. It is now opening the model up after a limited release to cybersecurity companies.

Google’s pitch promises “frontier performance on long-horizon agentic coding, enterprise knowledge work, and scientific reasoning, and a 1M output token limit.” In its launch post, Google says its internal teams have seen Carbon run for days on multi-step tasks such as root-causing and patching subtle bugs across large codebases. Google still points most work elsewhere, Engadget reports: Argon remains the better choice for most ordinary tasks and even the majority of software engineering tasks, and Carbon is meant for multi-step coding, scientific research and other complex work.

The Vertex AI billing catalog shows a split by tier. OfficeChai reported that StatsWire found 90 billing SKUs for Carbon. Online global pricing was listed at $4 input, $20 output and $0.20 for cache reads per million tokens, and Batch and Flex at $2, $10 and $0.10. The outlet also says billing listings are not an official product announcement.

The online figures match what Argon was set to cost after its introductory period. OfficeChai says Argon launched restricted on September 30 at $2/$10, and Google said at the time that pricing would rise to $4/$20 once the model opened more broadly.

Outside Google Cloud, Engadget lists Carbon in Google AI Studio and Gemini for business. Google AI Ultra subscribers, who pay $100 a month, get access in the Gemini app, Search and Antigravity, and 10,000 scientists are getting free access for six months.

Carbon launched at No. 1 on the Artificial Analysis Intelligence Index v5.0, with a score of 59. According to OfficeChai, Anthropic’s Fable 5.2 has since taken that spot.