Organizations gain agentic coding and 40+ integrations with Google's Gemini 4 Argon for Pro subscribers

Google has made its Gemini 4 Argon model available to subscribers of its Pro and Ultra tiers through the main Gemini application. According to a blog post from Google, this release delivers improved reasoning capabilities and enhanced performance for complex tasks. The model is designed to answer questions more directly, reducing unnecessary hedging and repetitive phrasing in its responses.

What we know

Argon provides major improvements for software development workflows, specifically through the integration of Google Antigravity’s agentic coding interface within the macOS app. Users can toggle between chat and task modes in the main window, allowing the model to perform actions across Google apps even when the user is not actively present at their computer. The system now connects to more than forty third-party applications, such as OpenTable and Shopify, enabling users to order groceries, book dining reservations, or manage business design tasks from the chat interface.

According to Google’s engineering team, Argon is being used by researchers and security specialists to advance work in science, mathematics, robotics, and cyber defence. The model ties for first place on the CWE-Bench benchmark for vulnerability remediation, achieving a score of sixty-eight per cent. Google states that Argon is also its most resilient model against indirect prompt injection attacks, a common technique used to bypass AI safety measures.

The model has assisted in significant software engineering projects, including helping the Chrome team refactor hundreds of files within the Chromium codebase. This effort separated core webpage handling from a legacy folder containing approximately six million lines of code. The changes aim to allow webpages to move fluidly between tabs and side panels without reloading, which Google says will eliminate a specific category of browser crashes for billions of users.

In the scientific sector, researchers used Argon within the Antigravity interface to optimise the xTB open-source quantum chemistry package. This tool predicts molecular structures and screens drug candidates. The optimisations made the package up to two point three times faster at predicting structures and one point eight times faster at conformer searches. These changes have been contributed to the official xTB repository on GitHub.

Additionally, the team behind Google’s open-source Agent Development Kit used Argon to reduce framework overhead. Working autonomously overnight, the model identified optimisations that cut in-memory session preparation time by forty per cent in long sessions and sped up cold imports by twenty-one per cent. These updates are now live in the google/adk-python repository.

Defender's checklist

• Verify that AI tools used for coding are restricted to approved internal repositories.
• Monitor for indirect prompt injection attempts in any integrated third-party applications.
• Review access controls for agentic features that can act without direct user supervision.
• Ensure that sensitive data is not processed through external app integrations like Shopify or Adobe.

The background in plain language

Agentic coding refers to AI systems that can execute complex software tasks autonomously, rather than just generating code snippets for human review. These systems can read, modify, and commit code changes independently. Indirect prompt injections occur when malicious data embedded in external inputs tricks an AI model into performing unintended actions. Teams usually restrict AI access to production environments and validate all automated code changes before deployment to mitigate these risks.

What to do and how to stay safe: Google

• Review which third-party applications your AI tools can access and disable unnecessary integrations.
• Monitor logs for any autonomous actions taken by AI agents outside of standard working hours.
• Validate all code changes generated by AI models before merging them into your main codebase.
• Ensure that sensitive data is not shared with AI models that connect to external services.

Step-by-step guide: AI Governance for Small Teams: Practical Controls and Cost

General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Who can access Gemini 4 Argon?

The model is currently available to subscribers of Google AI Pro and Google AI Ultra tiers through the Gemini app.

What new coding features does Argon include?

It introduces Google Antigravity’s agentic coding interface directly into the Gemini macOS app, allowing for more autonomous software engineering tasks.

Does Argon connect to external applications?

Yes, it connects to more than forty apps, including Instacart, OpenTable, Adobe, and Shopify, to help users complete tasks outside the Google ecosystem.