Google has opened Gemini 4 Argon, its most powerful artificial intelligence (AI) model, to businesses and developers worldwide, making it available through Google Cloud, Google AI Studio and the Gemini API from today.
The broader release follows a restricted launch on 30 September, when access was limited to vetted cybersecurity organisations through Google's Fairwind Program.
What Argon is
Argon sits at the top of Google's fourth-generation Gemini family and is the company's flagship frontier model, the industry term for the most capable AI a lab has produced at any given moment.
The race to lead that category is currently fought between three American laboratories: Google DeepMind, Anthropic and OpenAI, which fields GPT-6 Astra, while Anthropic offers Claude Opus 5.5.
Google says Argon is built for long, multi-step tasks that previous models could not sustain, including complex software engineering, legal and financial knowledge work, and cybersecurity defence.
What it can do
On DeepSWE v1.1, a benchmark for real-world coding performance, Argon scored 77.9%, against 74.2% for Claude Opus 5.5 and 74.1% for GPT-6 Astra.
On Google's broader benchmark table, Argon leads outright in 13 of 19 categories, with rivals ahead on the remainder across coding, science and computer-control tests.
Inside Google, engineering teams using Argon in Antigravity, the company's internal development environment where Argon is now the default model, increased agentic code submissions by more than 35% while cutting rollbacks by nearly a third.
The Chrome team used Argon to complete an architectural refactor across hundreds of files in Chromium's 40 million-line codebase, finishing in weeks what had been projected to take months.
Argon can handle up to 1 million output tokens per session, a token being roughly three-quarters of a word, compared with 64,000 for the previous generation.
Cybersecurity layer
Security is a distinct selling point, with Argon tying for first on CWE-Bench v1, which measures how effectively an AI can identify and fix software vulnerabilities, with a score of 68%.
The model also leads the Gray Swan Indirect Prompt Injection benchmark, reducing the success rate of attempts to hijack its behaviour to 0.7%, which matters when the model is processing emails, documents or web pages that may carry hidden malicious instructions.
Trusted cybersecurity organisations can still access Argon through the Fairwind Program with its defensive guardrails removed, allowing full capability testing unavailable to standard users.
Pricing and access
Argon launches at an introductory rate of $2 per million input tokens and $10 per million output tokens, with cached input priced at 95% off, until 30 November 2026.
Standard pricing of $4 per million input tokens and $20 per million output tokens takes effect after that date.
The model is available now via the Gemini platform under the model ID gemini-4-argon, in Google AI Studio and through Gemini for Business on web, desktop and mobile.
