Google launches Gemini 4 Argon globally for complex tasks

Bethlehem - Ofoq News - Google launched the “Gemini 4 Argon” model globally for developers and companies via the “Gemini API,” and also made it available to “Google AI Pro” and “Ultra” subscribers through “Gemini,” “Google Antigravity,” and “AI Mode” in the search engine. The company says the model was designed to handle complex, multi-step tasks in areas including programming, finance, law, cybersecurity, and science, and that it achieved advanced results in a number of evaluation tests across these areas.

According to Google, Argon ranked first on the Vals Index, which measures the performance of models in sectors including finance, programming, tax, and law. The model is available at an introductory price of $2 per million input codes and $10 per million output codes, with a 95 percent discount on cached input codes.

Increased productivity of programming teams

Among the most prominent applications presented by Google is the use of the model within its engineering teams. The company says that Argonne, through Antigravity, has helped increase agent-based code transmissions by more than 35 percent, while reducing code rollbacks by about a third.

In another example, the Chrome team used the model to help separate basic web page functionality from an old folder containing about 6 million lines of code within the Chromium database, which is about 40 million lines in size. According to the company, the model contributed to completing work that was expected to take months within weeks.

Applications in business and research

During the early access phase, companies including Deloitte, LexisNexis, Amcbeans Healthcare, and Isomorphic Labs used the model for tasks including in-depth research, gathering information from documents, analyzing long-term data, and reasoning across large codebases.

Google says that Argonne combines reasoning capabilities, visual understanding, expanded knowledge, and writing in multiple fields, with a focus on accuracy in complex tasks. The company also indicated that the model is being used to discover critical security vulnerabilities and proactively address them, within the range of uses that its teams and customers participating in early access are testing.

Applications in science

The use of “Argon” also extends to scientific research projects. Google says the model helped raise the productivity of quantum error correction in the AlphaQubit project to nearly double. She also pointed out its use in arriving at new mathematical solutions related to nuclear fusion research in cooperation with a professor at the University of Maryland, in addition to applications in life sciences and other research fields.

Abilities in robotics and spatial understanding

In the field of robotics, Google says that the model focuses on understanding three-dimensional spaces and visual concepts within applications related to physical artificial intelligence. Argonne scored 87 percent on the “Embodied Reasoning - ERQA” scale, which measures reasoning abilities related to visual and spatial understanding. According to the company, these capabilities rely on a combination of mathematical and logical reasoning and visual understanding, in addition to other uses of the model in programming, analysis, and scientific research.