Google Announces Gemini 4 Argon Model for Complex Workflows
Focused on software engineering, enterprise operations, and cybersecurity, the artificial intelligence model Gemini 4 Argon has been announced with a 1 million output token limit.
Google has introduced Gemini 4 Argon, a new artificial intelligence model developed for complex, long-running workflows such as software engineering, finance, legal, and cybersecurity. The model is initially being rolled out to trusted cyber defenders under the Fairwind Program.
Advanced Output Capacity and Pricing
The Argon model was designed to support complex and prolonged workflows requiring deep reasoning. Accordingly, Google increased the model's output token limit from the previous 64,000 level to a capacity of 1 million tokens.
The model will launch at a starting price of $2 per million input tokens and $10 per million output tokens. A 95% discount will be applied for cached input tokens.
Successes Achieved in Professional Benchmarks
Argon has been used in internal workflows for quantum algorithmic optimization, data center memory efficiency, and large-scale C/C++ to Rust codebase migrations, including the Fuchsia Zircon kernel exceeding 800,000 lines.
Setting a new benchmark at 77.9% on DeepSWE v1.1 in professional evaluation tests, the model took the lead in the Vals Index—which measures economic impact in finance, coding, legal, and tax domains—and ranked first in AutomationBench with a score of 51.3%.
Cybersecurity and Security Measures
Capable of autonomously finding, verifying, and patching critical software vulnerabilities in cybersecurity defense, Argon has begun to be utilized by Wiz as part of the Scan for Good initiative.
Prior to broader access, Google is enhancing frontier safety measures in four key areas: abuse prevention, prompt injection attacks, alignment monitoring, and system hardening in sandbox environments.