Google launches 'Gemini 4 Argon' and describes it as its most advanced model

Google unveiled its new model, 'Gemini 4 Argon', describing it as its most advanced model to date, to a select group of its cybersecurity partners on October 1. The company stated that the model allows an output limit of up to one million tokens, expanding the volume of output the model can generate in a single request.

Google explained that the rollout is taking place in phases and that the model is undergoing a U.S. security evaluation prior to its official launch, a practice that has become common among major tech companies before releasing advanced models. The company has not announced a specific date for when the model will reach regular users, as distribution begins with select partners.

Google said that the model sets a record in real-world software engineering tasks, ties for first place in cybersecurity benchmarks, and tops a test dedicated to professional tasks. These figures were released by the company itself and have not yet been independently verified, making comparison with competing models a matter requiring third-party testing.

The launch comes at a time when major tech companies are competing to deliver more powerful and faster models, amid growing demand for artificial intelligence tools in coding and security analysis. Users in the Arab region are eagerly anticipating the model's arrival on public platforms, as advanced releases typically take longer to reach regional markets compared to the U.S. market.

What do these terms mean?

Generative Model: An artificial intelligence system that learns from massive amounts of data and then produces text, code, or images based on a user's request. Token: The smallest unit of text processed by a model, used to measure input and output size; thus, a limit of one million tokens means the ability to handle very long texts in a single request. Benchmark test: A standardized set of tasks used to compare model performance against one another, though results remain tied to who conducted the test and how it was set up. Pre-launch security evaluation: A review conducted by the company or another entity to ensure the model does not generate dangerous outputs before making it available to the public.