Google Launches Gemini 4 Argon, Targeting Rivals With Competitive Pricing and Benchmark Claims
Google says Argon beats GPT-6 Astra and Claude Fable 5.1/Opus 5.5 across coding, cybersecurity, science, finance, and law benchmarks, while internal use halved YouTube comments compute. It launched October 13 at $2/$10 per million tokens, with rates doubling at November’s end.
Google released Gemini 4 Argon on October 13, 2026, marking the company’s first major frontier AI model launch since Gemini 3 in late 2025. The release comes after months in which models from Anthropic and OpenAI left Google trailing its competitors, and follows internal shakeups that raised questions about the company’s standing in the AI race.
Google says Argon delivers “frontier reasoning and improved performance across complex workflows” and claims it outperforms rival models — including OpenAI’s GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 — across benchmarks in coding, cybersecurity, science, finance, and law. The company also says Argon reduces hedging, repetition, and clichés common in AI chatbots.
Google reports it is already using Argon internally for tasks such as identifying and patching cybersecurity vulnerabilities, cutting the computing power required for YouTube’s comments section by half, and debugging the drone physics simulator for Alphabet’s Wing delivery service.
Argon is rolling out to Pro and Ultra subscribers in the Gemini app and AI mode in Search, as well as to Ultra subscribers on Google’s agentic development platform, Antigravity. Availability for Plus subscribers or free users has not been confirmed.
On pricing, Google is launching Argon at $2 per million input tokens and $10 per million output tokens — roughly one-fifth the cost of GPT-6 Astra and Anthropic’s Fable 5. Those introductory rates are set to double at the end of November, though Google says Argon will remain cheaper than frontier rivals on a per-token basis even after the increase. The company notes, however, that per-token pricing does not always translate directly to lower real-world costs, as models may use different numbers of tokens for identical tasks.
The launch follows a limited early release to a select group of “trusted cyber defenders” for risk testing, consistent with recent frontier model rollouts. Argon arrived one day after Meta announced its own frontier model, Muse Wonder Full, suggesting the competitive dynamics among leading AI developers remain fluid.