Google officially introduced Gemini 4 Argon on Wednesday, positioning its new flagship AI model as a frontier system for coding, office work, and cybersecurity defense. The launch comes one week after Anthropic's Claude Opus 5.5 and one day after OpenAI's GPT-6.1 Sol, underscoring the intensifying race among American AI labs.
On the DeepSWE v1.1 software engineering benchmark, Argon scored 77.9%, ahead of Claude Opus 5.5 at 74.2%, GPT-6 Astra at 74.1%, and Claude Fable 5.1 at 67.4%. Google's own comparison table showed Argon leading in 12 of 18 benchmarks, tying one and trailing in five. The model can also produce up to 1 million output tokens per reply, a major jump from 64,000 for previous versions.
Cybersecurity is the headline feature. On Gray Swan's Indirect Prompt Injection benchmark, Argon posted a 0.7% attack success rate, better than Claude Opus 5.5 and Claude Fable 5.1 at 1.0%, and far ahead of GPT-6 Astra at 8.5%, Grok 4.6 at 51.8%, and Kimi K3 at 52.7%. Google is first granting access through its Fairwind Program, a limited cyber defense initiative launched September 2 with more than 650 partners including governments and critical infrastructure operators. This early release ships without cyber guardrails, allowing vetted defenders to use offensive capabilities to identify and patch vulnerabilities before wider availability.
Google also reported that Argon scored 70.9% on the internal Wiz Penetration Test Benchmark, up from 58.2% for Gemini 3.8 Flash Cyber. Security firm Wiz used Argon to find a critical flaw in healthcare software that earlier frontier models missed.
Pricing is aggressive: introductory rates are $2 per million input tokens and $10 per million output tokens, undercutting Anthropic's high-end model at $10/$50 and OpenAI's comparable premium pricing. Standard rates will later rise to $4/$20. The phased rollout will continue with paid API customers and Google AI Ultra subscribers, followed by enterprise and general public access.
Google is participating in the U.S. government's voluntary pre-release model review process, and the launch coincided with President Trump unveiling a voluntary, penalty-free AI accord that Google's leadership signed. Alphabet's stock remained flat after the announcement, following a summer in which Alphabet shares fell about 4.4% after the company skipped the promised Gemini 3.5 Pro release.
Internally, Google says Argon is already being used for coding, research, and writing. One team used Argon agents to find memory savings across Google data centers, and the model is helping rewrite legacy code into Rust, including a system with more than 800,000 lines of code.
