Google unveils flagship model Gemini 4 Argon

On September 30, 2026, Google unveiled its new flagship model, Gemini 4 Argon. The announcement was published on the corporate blog, and the model itself is described as built for long-running work on complex tasks where holding context and carrying out multi-step reasoning matter.

The first to get access will be trusted cybersecurity specialists: the rollout is going through the Fairwind program. The company calls the approach phased and separately notes its participation in the voluntary pre-release procedure for model access run by the U.S. government.

The stated areas of use include software development, legal and financial work, and defense against cyberattacks. The output token limit was raised to 1 million from the previous 64 thousand, so the model can produce long chains of reasoning in a single pass. On the DeepSWE v1.1 benchmark, which evaluates real-world software development tasks, the model scored 77.9%.

Ahead of the broad launch, safety measures are listed: refusing harmful requests tied to cyberattacks and dangerous materials, resilience to indirect prompt injections, and monitoring of the model's behavior. Access is still limited; widening it to developers, companies, and ordinary users is planned for later, starting with paid API customers and Google AI Ultra subscribers.

For the introductory period, the price is given as $2 per million input tokens and $10 per million output tokens; after that, $4 and $20 respectively are stated.

Image:

Translation editor: Салтанов А.