{"id":14469,"date":"2026-10-15T06:00:55","date_gmt":"2026-10-15T06:00:55","guid":{"rendered":"https:\/\/krikis.lt\/google-gemini-4-argon-naujasis-modelis\/"},"modified":"2026-10-15T06:00:55","modified_gmt":"2026-10-15T06:00:55","slug":"google-gemini-4-argon-naujasis-modelis","status":"publish","type":"post","link":"https:\/\/krikis.lt\/en\/google-gemini-4-argon-new-model-2\/","title":{"rendered":"Google Unveils Gemini 4 Argon, an Advanced Model for Complex Coding, Cybersecurity, and Research"},"content":{"rendered":"

\u201eBusinessLine reports that Google has begun offering a new artificial intelligence model, Gemini 4 Argon, globally. The model is aimed at developers, enterprises, and select subscribers, and is designed to solve complex tasks in software development, finance, law, cybersecurity, and research. The model is available through the Gemini API, AI Studio, and Google Antigravity platforms, and Google AI Ultra and Pro subscribers can use it directly in the Gemini app and in AI mode in search.<\/p>\n

According to a Google blog post, the introductory price of $2 per million input tokens and $10 per million output tokens is topped up with a 95% discount on accumulated input tokens. This allows early adopters to test the model without incurring significant costs. Google also says Argon ranked first in the Vals Index, which measures the performance of AI models across a range of economic sectors, including finance, coding, law and tax.<\/p>\n

Coding and tooling improvements<\/h2>\n
\"Development<\/figure>\n

\u201eGoogle engineers working with Antigravity reported that using Argon increased agent-based code submissions by more than 35% and reduced code rollbacks by nearly a third. This shows that the model is able to efficiently handle long, multi-step tasks by combining logical reasoning with the use of multiple tools. For example, the Chrome team used Argon to perform an architectural refactoring of hundreds of files in a matter of weeks, a task that Google says would have previously taken months.<\/p>\n

\u201eGoogle CEO Sundar Pichai highlighted the model\u2019s ability to create complex animations from a simple Pixel CAD file at the X platform, from a 3D model to lighting, camera movement and sound. Pichai said Argon ran 1,000 iterations to test and refine the code, demonstrating a strong understanding of using multi-step tools.<\/p>\n

Research and cybersecurity<\/h2>\n
\"Cybersecurity<\/figure>\n

\u201eGoogle also highlighted Argon\u2019s applications in scientific research. The model helped researchers identify new enzyme designs that later demonstrated catalytic activity in the lab. In cybersecurity, Argon has been used to detect and fix vulnerabilities in government and healthcare systems. The company has implemented additional protections to combat malicious query types and hidden instructions in web pages, emails, and code. According to Google, this type of prompt-injection attack had a success rate of 0.7 % in fifteen adaptive tests.<\/p>\n

Availability and future plans<\/h2>\n

While initial availability was limited to AI Ultra and Pro subscribers, SiliconANGLE reports that the model is now available to Gemini app users on those subscription plans. It also introduces a new \u201etasks\u201c mode feature that allows the model to continue working when the user is away from the computer and provides continuous progress updates. This feature allows for the model to be linked to up to 40 different apps, including Instacart and OpenTable, expanding the model\u2019s usage scenarios.<\/p>\n

\u201eGoogle also announced that up to 10,000 researchers will receive six months of free access to the highest level of usage on the Antigravity platform if they meet the selection criteria, demonstrating the company's commitment to promoting scientific research and developing a community ecosystem.<\/p>\n

Conclusions<\/h2>\n

Gemini 4 Argon is not only a technical breakthrough, but also a practical tool that can transform programming, security, and research processes. Its ability to perform long, multi-step tasks, integrate with a variety of tools, and offer competitive pricing makes it attractive to developers, enterprises, and researchers alike. Google\u2019s plans to expand access and provide free access to researchers indicate that the model will become an important part of the AI ecosystem.<\/p>\n

Sources<\/h2>\n