Google Announces Gemini 4 Argon

Google promised Gemini 3.5 Pro earlier in the year, but instead spent the summer introducing various smaller Flash models. Now, according to an <a href="https://arstechnica.com/google/2026/09/google-announces-gemini-4-argon-ai-model-but-you-cant-use-it-today/">Ars Technica</a> report, the company is returning to the AI frontier with the announcement of <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/?utm_source=deepmind.google&utm_medium=referral&utm_campaign=gdm&utm_content=">Gemini 4 Argon</a>. Google claims this new artificial intelligence model delivers industry-leading performance across coding, knowledge work, and cybersecurity.

Although general consumers and enterprise customers cannot use the model just yet, Google notes that internal engineers are already utilizing Argon extensively. Internal applications include using fleet-wide telemetry data to save memory across data centers and migrating massive C/C++ codebases to Rust across the company.

Benchmarks and Output Limits

Google has shared a series of benchmarks to support its performance claims. On the DeepSWE v1.1 software engineering benchmark, Gemini 4 Argon reaches a score of 77.9 percent, outperforming several competing models. Furthermore, Google points to Argon's performance in the economic analysis Vals Index test as evidence of its capabilities across long-horizon tasks.

Because the model remains in limited testing, API pricing has not been disclosed. However, Google confirmed that Gemini 4 Argon will support an output limit of 1 million tokens, representing a massive increase from the 64,000-token limit found in previous Gemini models. This expansion aims to help users tackle larger, more daunting tasks in a single step.

Cybersecurity Focus and Phased Release

The primary focus for Gemini 4 at this stage is cyberdefense. Due to its scale, Google is rolling out the model via a phased release starting with a small group of trusted testers and partners in the company’s Fairwind Program. Partners like Wiz are already using Argon to identify critical security vulnerabilities in hospital systems globally.

To address concerns regarding model safety and potential misalignment following summer hacking incidents, Google integrated systems that monitor the model's chain-of-thought. These mechanisms can halt the model if it goes out of bounds, highlighting the role of reasoning transparency in frontier model development.

Availability and Future Outlook

Google has not provided a strict timeline for when Gemini 4 Argon will become broadly available to enterprise and consumer customers. Following initial cybersecurity testing, general availability will roll out first to paid API users and Google AI Ultra subscribers.

Sources

• Ars TechnicaGoogle announces Gemini 4 Argon AI model, but you can't use it yet