Google unveils Gemini 4 Argon, which it calls its most powerful Gemini model

Alphabet, Google’s parent company, has unveiled its new Gemini 4 Argon artificial intelligence model. According to the company, the model was designed to perform various complex tasks, including programming, research and writing. However, Google emphasizes that Gemini 4 Argon has particularly strong cybersecurity capabilities.

Argon is not currently available to the general public. The model is being offered only to select cybersecurity partners through Google’s security initiative, the Fairwind Program. Argon was specifically designed for defensive cybersecurity tasks. Google says the model can independently detect, verify and remediate critical software vulnerabilities.

Gemini 4 Argon also delivers strong results in programming and engineering tasks. Google employees are already using the model in their daily work, including to detect coding errors and migrate large codebases to new systems.

In one example from Google, Gemini 4 Argon optimized the Rust version of the libgav1 video decoder. According to the company, the resulting program ran 2.7 times faster than the previous Rust version, with no change in video output quality.

Another advantage of the model is its ability to analyze visual information. Argon can examine the content of long videos and analyze charts and other visual materials. Google says the model was designed to reason deeply while carrying out complex, long-running, multistep tasks.

Google has also increased the model’s output limit from the previous 64,000 tokens to 1 million tokens. This allows Gemini 4 Argon to work with larger volumes of data and carry out long-running, complex tasks with broader context.

The company claims Gemini 4 Argon outperformed OpenAI’s GPT-6 Astra, as well as Anthropic’s Fable and Opus models, in a number of artificial intelligence tests. The model achieved a 77.9% score on the DeepSWE v1.1 programming test. This benchmark evaluates performance on long-running, real-world software engineering tasks. Google bases these results on various benchmarks and says Argon currently leads Vals’ artificial intelligence model index.

Gemini 4 Argon’s initial API pricing is $2 per 1 million input tokens and $10 per 1 million output tokens. Cached input tokens are priced 95% below the standard input rate.

Google has recently been working to strengthen its position in the artificial intelligence market with the Gemini family. The company announced in August that the Gemini app had surpassed 1 billion monthly users. For comparison, OpenAI also announced that ChatGPT had reached 1 billion monthly users.