Google began rolling out Gemini 4 Argon to the public on Tuesday, limiting access to paying users. Argon is the first model in a new Gemini 4 family, and Google says it excels at coding, cybersecurity and knowledge work in finance and law.
Google announced on Tuesday that Gemini 4 "Argon" is rolling out to Gemini app users and developers. The company is positioning Argon as its new workhorse model. Tulsee Doshi, Gemini product lead at Google DeepMind, said Argon would be the first of a family of new Gemini models rolling out over time.
Paid access only, with API pricing from $2
Access requires a subscription. Argon is rolling out to Pro and Ultra Gemini app subscribers, who pay at least $19.99 per month, and those users will also see it in AI Mode while searching on Google. Developers will also need to pay for the model.
On the developer side, API access launches at an introductory price of $2 per million input tokens and $10 per million output tokens. Cached inputs receive a 95% discount, according to Crypto Briefing. Google said Argon supports a unified input and output limit of one million tokens.
Google cites internal gains and benchmark results
Google first unveiled Argon last month, but only a set of trusted partners could test it and address cyber vulnerabilities. Google has since tested the model internally with employees for several weeks, and it said Argon helped engineers increase the production of AI-generated code by 35%.
According to the company, Argon ranks first on the Vals Index, which measures performance across sectors including finance, coding, legal and tax work. Box said Argon increased accuracy by 14% compared with Gemini 3.1 Pro on enterprise knowledge tasks. Crypto Briefing noted that these results reflect evaluations conducted by Google and its early customers.
Competition and safeguards
The launch comes amid competition between the big AI companies. Meta said on Monday that its new Muse Wonder class of models would bring the company to the frontier, and last month OpenAI started rolling out GPT-6. Argon also gives Google a chance to reassert itself after it chose not to launch its previously promised Gemini 3.5 Pro.
Google said it has stepped up efforts to keep the model from misbehaving, using a system that tracks the model's chain of thought. Google also said Argon recorded a 0.7% attack success rate on the Gray Swan Indirect Prompt Injection benchmark across 15 adaptive attempts.
Sources: Business Insider, Crypto Briefing
Trading involves risk.
Risk Disclaimer
