Google introduces Gemini 4 Carbon model

Google has introduced a new flagship artificial intelligence model called Gemini 4 Carbon. According to the company, Carbon, which is built on the Argon model, is designed for efficient performance in areas such as programming, scientific research, and the analysis of data in various formats.

One of the key features of the new model is its ability to independently perform complex tasks over long periods. Google states that Gemini 4 Carbon was developed to carry out agentic tasks continuously for several days. This approach allows the artificial intelligence not only to answer a single question but also to independently manage multi-stage workflows.

In test results presented by the company, Carbon recorded higher performance metrics than a number of competing models. Specifically, in the Terminal Bench 4.0 (AA) agentic programming test, Gemini 4 Carbon achieved a score of 65.2 percent. According to the information provided in the text, the results for the Claude Opus 5.5 and GPT-6 Astra models were 59 percent.

Currently, Gemini 4 Carbon is in the pre-release testing phase. By November 2026, it is planned to be made available to select trusted companies. After the public launch, the model's pricing is expected to be the same as Argon: $2 per million input tokens and $10 per million output tokens. The cost for cached input tokens will be 95 percent cheaper.