Google Gemini 4 Argon is here, but few can use its most powerful AI yet
&imwidth=600&imheight=450&format=webp&quality=medium "Google Gemini 4 Argon is here, but few can use its most powerful AI yet")
Google Gemini 4 Argon is here, but few can use its most powerful AI yet
Google's Gemini 4 Argon is a new frontier AI model with a 1 million-token output limit and capabilities across coding, enterprise work and cybersecurity. It is initially restricted to trusted cyber defenders, with wider access planned for developers and consumers later.
Google has unveiled Gemini 4 Argon, its latest frontier AI model, but most users cannot access it yet. The model is initially being rolled out to a small group of trusted cyber defenders through Google's Fairwind Programme, as the company tests its safety systems before a wider launch.
The restricted release makes Argon unusual even in the fast-moving AI market. Google says the model is capable of handling complex, long-running tasks across software engineering, legal and financial work and cybersecurity. It also comes with a major technical upgrade: a 1 million-token maximum output limit, up from 64,000 tokens on previous Gemini models. That gives Argon considerably more room to reason through lengthy, multi-step tasks in a single run.
Why Google is keeping Gemini 4 Argon restricted
Google says the phased launch is necessary because frontier AI capabilities can create new safety and security risks. The company is participating in a US government voluntary process for pre-release access and plans to use feedback from early testers to improve its guardrails. Argon has been trained for cybersecurity defence and can autonomously identify, validate and patch software vulnerabilities, according to Google. The model is already being used internally. Google says Argon agents analysed data-centre telemetry and identified memory optimisations that could free more than 300 TiB of memory once deployed, with estimated total savings of 500 TiB to 1 PiB.
Gemini 4 Argon vs OpenAI and Anthropic
Google's published benchmarks show a mixed picture rather than a clean sweep. Argon scores 68.9 per cent on the Vals Index, compared with 67.0 per cent for Claude Opus 5.5. It also scores 77.9 per cent on DeepSWE v1.1, ahead of Opus 5.5's 74.2 per cent. But rivals still lead on some tasks. Claude Opus 5.5 scores 66.4 per cent on Terminal-Bench 4.0, compared with Argon's 57.4 per cent. GPT-6 Astra also leads Argon on several other evaluations, including FrontierSWE v2 and OSWorld-2.0. That means Google's results point to a highly capable model, but not one that dominates every workload.
Trending Stories
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
Gemini 4 Argon price and wider release
Google says Argon will initially cost $2 per million input tokens and $10 per million output tokens. Cached input is priced 95 per cent below the standard input rate. After the introductory period, prices rise to $4 and $20 respectively. For now, access remains limited to trusted cyber defenders. Google says broader availability will begin with paid API customers and Google AI Ultra subscribers before expanding further, although it has not given a specific public release date. The bigger question is therefore not simply whether Gemini 4 Argon is powerful. It is whether Google's combination of long-context reasoning, agentic capabilities and competitive pricing can turn that capability into a meaningful advantage once the model reaches more developers and businesses.
