Gemini 4 Argon available on Google Cloud | Google Cloud Blog
Short summary
Google released Gemini 4 Argon, a new frontier model designed to handle complex multi-step business tasks across software development, finance, legal work, and security. The model can maintain context over long documents, deliver precise results with verifiable citations, and execute autonomous work from a single prompt.
- Argon handles multi-step tasks that previously took hours or days, processing long documents with mixed text, tables, charts, and images into structured outputs with honest reporting of limitations.
- In regulated domains, Argon leads on benchmarks for precision: it tops the Vals Index for financial and legal work, scores 68% on CWE-Bench for security vulnerability remediation, and achieved 20% better task success than earlier models in ServiceNow's testing.
- Google's internal teams used Argon to cut code session prep time by 40%, reduce rollbacks by nearly a third, and compress months of architectural refactoring work into weeks on Chrome's 40-million-line codebase.
- Argon resists indirect prompt injection attacks at 0.7% success rate across 15 adaptive attempts and includes defensive security capabilities while blocking offensive tasks like pentesting.
Full summary
Google just released Gemini 4 Argon, a new model designed to handle the kind of work that typically eats up days or weeks—complex financial analyses, legal research, software builds, security audits. The key difference from previous versions: Argon can maintain context across long, multi-step tasks without losing track. It reads messy inputs (documents with mixed text, tables, charts, images) and produces clean, verifiable outputs. It also knows its limits—it'll tell you when it can't find something or needs confirmation before doing something irreversible. Google trained it specifically for precision in regulated work like finance and law, where calculations and citations need to be bulletproof.
The concrete wins are worth noting. Argon scored highest on the Vals Index, which measures financial and coding work. It leads on Harvey's legal benchmarks for research and M&A work. ServiceNow saw roughly 20% better task success compared to Gemini 3.8 Flash. On Google's own projects, Argon optimized Python code to cut session prep time by 40% and cold imports by 21%. Chrome's engineering team used it to complete months of architectural refactoring across Chromium's 40-million-line codebase in weeks. These aren't marketing claims—they're specific metrics from real deployments.
On security, Argon ties for first on CWE-Bench v1 with 68% accuracy for fixing vulnerabilities. More importantly, it's hardened against prompt injection attacks—the kind where hidden malicious instructions try to hijack the model's behavior. It scores 0.7% attack success on the Gray Swan benchmark across 15 adaptive attempts. Google's also releasing an ungated version to vetted security partners through their Fairwind Program, letting trusted defenders test the full range of capabilities without restrictions. The model is available now on Google Cloud.