Google has introduced its latest frontier model, Gemini 4 Argon, but the reception on Wall Street is tempered by a growing focus on personal AI agents. According to CNBC, investors are less impressed by raw model benchmarks and more interested in products that can autonomously handle multi-step tasks for everyday users. JPMorgan analysts noted that Google has struggled to turn steady product development into splashy consumer releases that capture sustained attention.
Cost and access are immediate concerns. Argon's introductory pricing is competitive, but Google says published rates will eventually double to $4 per million input tokens and $20 per million output tokens. Running agents continuously is expensive, and Google has yet to show how it will bundle models into affordable agent services. Argon is also not broadly available: access is limited to select cybersecurity defenders and enterprise cloud customers, with no public release date, which limits independent testing.
There is also a question about real-world performance. Bloomberg reported that some Google employees questioned the model's coding abilities despite strong benchmark results. Google disputes that characterization, telling CNBC that employees across the company have been testing Gemini 4 versions for weeks, with some receiving unlimited access. The disagreement highlights the gap between benchmark scores and practical usefulness as the industry races toward agentic AI.