Google launches Gemini 4 Argon AI model globally for developers and enterprises

Google has launched Gemini 4 Argon, rolling out the frontier artificial intelligence model worldwide across developer tools, enterprise platforms, and consumer applications.

The system is accessible immediately through the Gemini API in Google AI Studio and Android Studio, as well as inside Google Antigravity and AI Mode in Search. Subscribers to Google AI Pro and AI Ultra can access the model directly within the Gemini mobile and web apps. The release represents Google’s primary frontier launch since Gemini 3 Pro, skipping an interim 3.5 release.

For enterprise teams and India’s expanding developer ecosystem—increasingly supported by national digital infrastructure and large-scale tech adoption—the launch introduces high-end reasoning tools tailored for complex software refactoring, cybersecurity defence, and industrial robotics.

Developer pricing and token limits

Google has introduced the model with a temporary pricing cut running through 30 November 2026. The company is offering cached input tokens at a 95% discount during this promotional window.

The model features a unified 1-million-token output limit, significantly exceeding the roughly 128,000-token caps common among competing frontier models, allowing developers to generate and audit extensive codebases in single passes.

Benchmark performance and engineering benchmarks

According to Google, Argon holds the top spot on the Vals Index, an evaluation suite measuring technical performance across software engineering, legal compliance, taxation, and financial analysis. On the Embodied Reasoning benchmark (ERQA), which measures spatial problem-solving in three-dimensional environments, the system registered 87%.

Internally, Google deployed the model across its own operations before public release. Software engineers using Argon within the Antigravity development environment saw agentic code submissions rise by more than 35%, while code rollbacks dropped by roughly one-third. In one major infrastructure task, the model helped the Chrome team isolate core webpage rendering from a 6-million-line legacy segment inside Chromium's 40-million-line codebase in several weeks.

Scientific research and security deployments

Outside pure code generation, Google reported early deployments across science and cybersecurity. Cybersecurity partner Wiz deployed early builds of Argon to chain multi-step software vulnerabilities, closing an administrative bypass in healthcare software that held records for hundreds of thousands of patients.

In scientific research, Google said the model helped scientists discover broader counterexamples to Grad’s conjecture in stellarator plasma physics alongside University of Maryland researcher Matt Landreman. The system also assisted in the synthetic design of five functional serine esterase enzymes, which were subsequently validated in wet-lab tests.

To drive broader research adoption, Google announced an initiative granting 10,000 eligible scientific researchers six months of complimentary access to Antigravity’s highest usage tier.