Google Unveils Gemini 4 Argon for Complex Enterprise Workflows
The new frontier model features a 1 million token output limit and is currently being deployed to cyber defenders via the Fairwind Program.
Google has announced Gemini 4 Argon, a frontier model designed for long-horizon professional tasks including software engineering, cybersecurity, and enterprise knowledge work in finance and law. As Google first reported, the model is currently being rolled out to a select group of trusted cyber defenders through the company's Fairwind Program.
Technical specifications detailed by Google include a significant expansion of the output token limit to 1 million tokens, an increase from the previous 64K tokens. This expanded headroom is intended to allow the model to sustain deeper reasoning for multi-step problem solving. Google reports that Argon has achieved a new state of the art on DeepSWE v1.1 with a score of 77.9% and ranks first on Zapier’s AutomationBench with a score of 51.3%. The company also notes leading performance on the Vals Index, as well as the Vals Finance Agent v2 and Harvey’s Legal Agent Benchmark.
Internally, Google reports that Argon agents have autonomously identified memory optimizations across data centers, resulting in over 300 TiB of freed memory with total estimated savings between 500 TiB and 1 PiB. The model is also being used to migrate C/C++ codebases to Rust, including the Fuchsia Zircon kernel. In one instance involving libgav1, Argon agents produced a memory-safe video decoder that runs 2.7x faster than a previous Rust port.
Introductory pricing is set at $2 per million input tokens and $10 per million output tokens, while cached input tokens receive a 95% discount. The company stated it is participating in a voluntary process established by the U.S. government for model access prior to a general public release.