Google released its flagship Gemini 4 Argon model this week, touting benchmark scores that place it just behind Anthropic's top models. The company is rolling the model out first to cybersecurity partners while analysts wait to see how it performs once businesses deploy it widely.
The company has touted Gemini 4 Argon as outperforming top OpenAI and Anthropic offerings on some benchmarks.
Argon lands behind Anthropic on the leaderboard
The Artificial Analysis Intelligence Index, a composite benchmark score, places Gemini 4 behind only Claude Opus 5.5 and Claude Sonnet 5.5 among top labs. Tim Law, director of research for AI at IDC, said the model shows advanced reasoning on legal work, finance and other enterprise knowledge tasks, including long-running jobs.
Lian Jye Su, chief analyst at Omdia, said Google had been late to the cybersecurity and coding game, yet Argon has pushed the company back to the frontier, particularly in cybersecurity. Nick Patience, AI lead at the Futurum Group, said Argon has made Google competitive again, though it doesn't make the company the outright leader. As Law put it, Argon "opens up another competitive front for enterprise knowledge work with other leading frontier models."
Cybersecurity partners get the first access
Google plans to launch Argon in phases, starting with trusted cybersecurity partners while it works with the U.S. government on pre-release safety evaluations. Tulsee Doshi, Google's Gemini model product lead, told CNBC that starting the rollout this way builds confidence and gets a model built for cyber defense into defenders' hands sooner.
Security incidents at OpenAI and Anthropic have created an opportunity for Google, according to Su, who said the episodes let Google stake its claim as the trusted model provider for safe and secure AI operations.
Wider deployment remains the real test
Google said it is already using Argon internally to optimize memory at its data centers, freeing up hundreds of terabytes without buying additional hardware. Quantum computing researchers have also used the model.
But the company has given no timeline for a broader release. While Google highlights internal use cases, Law said the final proof will come once the model reaches enterprise production environments.
Source: CNBC
Trading involves risk.
Risk Disclaimer
