Google Releases Gemini 4 Argon Programming Model

Google DeepMind unveiled its flagship programming model, Gemini 4 Argon, designed for coding, legal and financial sectors, as well as cybersecurity. Its key feature is support for up to 1 million output tokens.

Internal tests at Google demonstrated the model's high efficiency. It optimized data centers, freeing up over 300 TB of memory, and rewrote the Zircon kernel from C/C++ to Rust.

• Data center optimization and freeing up over 300 TB of memory
• Rewriting the Zircon kernel code (800K+ lines) in Rust
• Accelerating video decoding in libgav1 by 2.7x by replacing SIMD code

The model was released without cyber restrictions for trusted security professionals. Using it, Wiz discovered a critical vulnerability in medical software that previous versions had missed.

Access to Gemini 4 Argon is currently limited to participants of the Fairwind program and the US government. For paid API clients and Google AI Ultra subscribers, the price will be $2 per million input tokens and $10 per output.