Google has announced Gemini 4 Argon, a new frontier AI model designed to handle complex reasoning, software engineering, enterprise knowledge work and cybersecurity defense. The model is initially being made available to a limited group of trusted cyber defenders through Google DeepMind’s Fairwind Program as the company takes a phased approach to wider deployment.
Gemini 4 Argon is designed for long-running workflows and can generate up to 1 million output tokens, a major increase from the previous 64,000-token limit. Google says the expanded context and output capacity allow Argon to work through complicated problems over much longer reasoning trajectories.
The model is already being used internally by Google teams for coding, research and engineering tasks. In one quantum computing application, Google says Argon helped researchers optimize a bottlenecking algorithm and exceeded a published baseline by 40%. Other Argon agents analyzed data-center telemetry and identified memory optimizations that could eventually free more than 500 TiB of memory.
Google is also using Argon for large-scale software migration projects. Agents have been involved in moving C and C++ codebases to Rust, including projects ranging from tens of thousands of lines to more than 800,000 lines of code. Google says one project involving the libgav1 video decoder resulted in a Rust implementation that was 2.7 times faster than an earlier Rust port while producing identical video output.
The company is positioning Argon as a strong model for enterprise work beyond programming. Google reports leading results across finance, legal research, tax, automation and long-video understanding benchmarks. On the DeepSWE v1.1 benchmark for real-world software engineering tasks, Argon achieved a score of 77.9%, according to Google.
Cybersecurity is another major focus of the new model. Gemini 4 Argon has been trained to identify, validate and patch software vulnerabilities. Google says it tied for first place on CWE-bench v1 with a score of 68%. Wiz is also testing Argon through its Scan for Good initiative to identify and remediate security risks affecting critical infrastructure.
Google says it is strengthening safeguards against misuse, prompt injection, misalignment and security risks before expanding access. The model is expected to reach developers, enterprises and consumers in phases, beginning with paid API customers and Google AI Ultra subscribers.
At launch, Gemini 4 Argon is priced at $2 per million input tokens and $10 per million output tokens, with cached input tokens receiving a 95% discount.
