Google has opened access to its new flagship model Gemini 4 Carbon, which, according to AIN.Capital, is intended for complex tasks in programming, cybersecurity, data analysis and scientific research. The company states that the model can execute multi-step tasks, verify its own results and collaborate with other AI agents.
Unlike the Gemini 4 Argon model, which is focused on everyday programming and information work, Carbon spends more computing resources on finding solutions and verifying results. The model can work on large software projects, find bugs in code, fix them and check whether the changes introduced new problems. In web application development, it can create applications, run them, analyze the output and independently correct detected errors, and it also works with 3D graphics and game development.
Google also presented the /swarm feature in the Antigravity development environment, which enables running multiple AI agents on the same task. The agents work in parallel, and the user can track their progress in real time. According to test results published by the company, Carbon took first place on CursorBench 4.0, Perplexity WANDR and Zapier AutomationBench, with a cost per task more than twice lower than the nearest competitor.
The model is available to developers via a paid API in Google AI Studio and Android Studio, and to companies through Gemini for business. The introductory API price is valid until November 30, 2026 and amounts to 2 dollars per million input tokens and 10 dollars per million output tokens, after which it increases to 4 and 20 dollars.
What this means for businesses
For companies developing software, the most interesting part is the combination of autonomous bug fixing and parallel operation of multiple agents, which can reduce the time needed for testing and code review. The introductory API price is valid until November 30, 2026, so companies planning a larger volume of use have a limited period to assess costs. Google emphasizes that Carbon is intended for the most complex tasks, while Argon remains the model for everyday workflows, which means that the choice of model will depend on the type of work. The test results cited by the company have not been independently verified, so they should be interpreted with caution.
This summary was written with AI based on the cited source and checked by automated rules. For exact details, read the original article.