Google has unveiled its new flagship AI model, Gemini 4 Argon. It is already being used internally for complex engineering tasks. For example, to find vulnerabilities or port large C/C++ projects to Rust. But the model is not yet available to ordinary users.
Google has unveiled its new flagship AI model, Gemini 4 Argon. It is already being used internally for complex engineering tasks. For example, to find vulnerabilities or port large C/C++ projects to Rust. But the model is not yet available to ordinary users.
This is reported by Google.
One of the most notable changes is the maximum response size. For Argon, Google has increased the output token limit from 64,000 to 1 million. The company hopes this will allow the model to work on a single complex task for longer without having to break it up into separate sessions.
Argon is already being tested by thousands of Googlers. In one such project, Argon rewrote 32,000 lines of code for the libgav1 video decoder in Rust and found a way to make the new version faster. Google says the decoder now runs 2.7 times faster than the previous version, but with the same performance.
Another big area for Argon is cybersecurity. Google says the model can independently find, inspect, and fix critical vulnerabilities. For select, vetted cybersecurity teams, the company plans to provide a version without standard cyber restrictions so they can use its full capabilities to protect systems.
There is no wide launch yet. Google will first test the security mechanisms and collect feedback, and then plans to open Argon to developers, businesses, and regular users. The first among them should be paid API customers and Google AI Ultra subscribers. Gemini 4 Argon will cost $2 for 1 million input tokens and $10 for 1 million output tokens. For cached input tokens, Google promises a 95% discount.
Previously, dev.ua wrote that Google has connected Gemini to Google Wallet. The AI assistant can find information from saved tickets and loyalty cards, as well as analyze the user's spending and summarize.
