Mountain View, 2 October 2026 – Google’s controlled rollout of its Gemini 4 model is placing a familiar artificial-intelligence debate back in focus: whether leading benchmark results translate into consistently superior performance for real users. The company has begun providing the model, internally known as Argon, to a small group of trusted cybersecurity partners and plans a wider release after testing, beginning with paying subscribers.
Google says Gemini 4 delivers leading results across a range of evaluations, including an advantage over OpenAI’s Astra on a benchmark measuring cybersecurity skills. The model has also been designed for long, complex tasks and can process up to one million tokens, equivalent to roughly 750,000 words. Its strengths are said to include multimodal analysis, natural communication, safety and the ability to interpret video metadata, capabilities that could support enterprise research, security work and large-scale document processing.
Unlock the Full Article
This article is exclusive to The Ledger Asia Subsribers / PAID members.
Already have an account? Log in here