Google has launched Gemini 4 Argon, its first major frontier AI model since Gemini 3 arrived in late 2025, and says it can compete with the industry’s leading systems. For now, access is limited to higher-tier subscribers, while most of the performance claims come from Google itself—the traditional frontier-model launch arrangement, in which the company supplies both the product and the very impressive report card.
Google says Argon offers “frontier reasoning and improved performance across complex workflows,” with less hedging, repetition, and chatbot cliché. The company also claims it outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across benchmarks in coding, cybersecurity, science, finance, and law. The supplied announcement does not include benchmark scores, so the breadth of the claims is clearer than their margin or independent significance.
Google says it is already using Argon internally to find and patch cybersecurity vulnerabilities, reduce by half the computing power needed for YouTube’s comments section, and debug the drone physics simulator for Alphabet delivery company Wing. Those are potentially useful applications, particularly if the reported efficiency gains hold up in practice. The announcement, however, does not provide further details about the tasks or how those results were measured.
The model is rolling out today in the Gemini app and Search’s AI mode for Google Pro and Ultra subscribers. It is also available to Ultra subscribers on Antigravity, Google’s agentic development platform. Google has not said when—or whether—Plus subscribers or free users will get access. The model may be front-of-mind for the companies competing to build the best AI, but availability still comes with a tiered-access gate.
Argon’s introductory API price is $2 per million input tokens and $10 per million output tokens. That is around a fifth of the cost of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models, according to the source. Prices are set to double when the introductory period ends at the end of November; even then, Google says Argon will remain significantly cheaper per token than its frontier rivals.
Per-token price, though, is not the same thing as the bill for a finished task. Models can use different numbers of tokens to do the same work, so a lower listed rate does not automatically mean lower overall costs. It is a useful opening price, not a universal coupon for AI computing.
Before the public rollout, Google gave Argon to a select group of “trusted cyber defenders” to test for potential risks, following a pattern used with other recent frontier models. The launch comes after months in which OpenAI and Anthropic drew attention with new models, amid internal upheaval that raised questions about Google’s ability—or appetite—to keep pace. Meta announced its own frontier model, Muse Wonder Full, a day before Argon’s arrival.
Argon gives Google a new contender, not a verdict on the race. The company’s claims and aggressive pricing make the launch worth watching; the missing benchmark detail and uncertain access for lower-tier users are reminders that a polished model announcement is not the same thing as a demonstrated lead. The frontier keeps moving. The sales pitch, as ever, arrives right on time.
Sources: Glitch Please! and The Verge — https://www.theverge.com/ai-artificial-intelligence/1009862/google-releases-gemini-4-argon — Reporting by Robert Hart; source material for Google’s claims, product availability, pricing, and launch context.
Generated by AI
