Google has begun the rollout of its flagship Gemini 4 Argon AI model to select cybersecurity partners, with plans to expand access to paid subscribers. While the company reports that the model outperforms OpenAI Astra in security benchmarks, internal sources suggest it struggles with practical programming tasks, particularly in frontend development.
The development of such systems is costly, with training expenses estimated at $400 million, excluding high engineering salaries. Concerns have been raised regarding the model's large size, which may impact Google's profitability. Critics suggest the company has focused too heavily on benchmark optimization rather than real-world performance.
Despite internal skepticism, Google leadership remains optimistic. The company notes that its AI-powered search and chatbot services have surpassed 1 billion users. Gemini 4 is designed for complex tasks in software development, finance, and law, capable of generating up to 1 million tokens, or approximately 750,000 words, in a single session.
The company has faced significant leadership changes, including the transition of DeepMind's leadership to Koray Kavukcuoglu. Google previously canceled the release of Gemini 3.5 Pro in June, a move that reportedly affected its market position. The team continues to compete against rivals like Anthropic and OpenAI, which are increasingly focused on autonomous AI agents.
