Google had promised a major model release for June, but the expected Gemini 3.5 Pro never arrived as the company continued shipping smaller Flash models. The long delay fed a perception that OpenAI and Anthropic had pulled ahead in the most demanding coding, research and cybersecurity work.¹

On Wednesday, Google unveiled Gemini 4 Argon, presenting it as its answer. CEO Sundar Pichai called it a model with “frontier performance in complex workflows, cyber defense and software engineering,” adding that Google teams were already using it extensively.Sundar Pichai (sundarpichai)² The company says Argon has helped with large-scale code migrations and data-center optimization, freeing more than 300 tebibytes of memory without additional hardware.³

But Google is not opening the doors just yet. Argon is initially available to vetted government and cybersecurity partners through its Fairwind Program while the company works through U.S. pre-release safety processes. Tulsee Doshi, Google’s Gemini product lead, said the phased approach both builds confidence and puts a powerful defensive tool “in the hands of defenders as soon as possible.”⁴ Google says it is strengthening safeguards against misuse, prompt injection and misalignment before broader access.⁵

The early scorecard gives Google ammunition: Argon has claimed top or near-top results across coding, enterprise knowledge-work and cyber benchmarks, with one widely shared result putting it first on the Vals Index at 68.9%.Demis Hassabis (demishassabis)⁶ Analysts, however, draw a sharper line between a benchmark comeback and market leadership. IDC’s Tim Law said the final proof will come only once the model reaches enterprise production environments; Google has offered no timetable for that wider release.⁷

That leaves Argon in an unusual position: Google’s loudest frontier claim in months, and a model most prospective users still cannot try.