Google Releases Gemini 4 Argon Globally to Paying Users

Google has released its Gemini 4 Argon artificial intelligence model globally to paying users as of Tuesday. The model is accessible through the Gemini app, Google Search, and the developer API. This launch follows earlier reporting by Threat Times on Google's expansion of AI-driven consumer tools, such as the Universal Cart introduced in October 2026.

Tulsee Doshi, the Gemini product lead at Google DeepMind, described the new model as a powerhouse. She told Business Insider that Argon is the first in a new family of models. Google claims the system ranks first on the Vals Index, a benchmark that measures performance in finance, coding, legal work, and tax preparation tasks.

Pricing and Access Tiers

Developers can access the model through the API at a discounted launch rate. Google said on Tuesday that the cost is $2 per million input tokens and $10 per million output tokens. These prices are temporary launch discounts. The standard rates for Argon are $4 per million input tokens and $20 per million output tokens. Google stated the discount will run for at least one month but did not specify an exact end date.

Consumers need a Pro subscription to use the model in the Gemini app or Search. The Pro plan starts at $19.99 per month. Google moved free users to its smallest model four days ago. More advanced capabilities, such as the Antigravity agent tool, require an Ultra subscription starting at $99.99 per month.

Early Testing and Safety Metrics

The first users of Argon were not developers but a small group of cyber defenders. Google provided these users with the model without its standard guardrails through a program called Fairwind. That group is now growing to hundreds of partners. Wiz, one of the partners, used the model to scan critical infrastructure. Wiz said it found a bypass exposing a master key in healthcare software during this testing phase.

Google provided internal safety metrics to reassure users about the model's robustness. The company said external bio red teaming unlocked under 1% of the sub-tasks it tests for. Shortcut attempts in coding fell to 2%. Prompt injection attacks succeeded in only 0.7% of cases. Google also runs a monitor that reads the model’s reasoning in real time to catch errors before they finish executing.

Robotics and Simulation Capabilities

Google is positioning Gemini 4 Argon as a single model for the full robotics pipeline. The system controls robots, generates interactive 3D training environments from photos or sketches, and grades training videos. According to The Decoder, the model scores 87 percent on Google's own ERQA benchmark. However, published example prompts reveal that the demos still require careful human guidance and detailed instructions.

In one demonstration, a Boston Dynamics Spot robot used Argon to build a virtual copy of a trash bin. The robot practiced balancing on three legs inside the simulation before attempting the task on real hardware. Two extra AI agents verified the positioning of the foot and gripper before the robot dropped a soda can. This approach aims to move toward physical AGI, a long-standing goal for Google and DeepMind.

What should you do?

Organizations evaluating new AI models should review the specific safety metrics provided by vendors, such as prompt injection success rates. Developers using discounted API access should monitor for changes in pricing and rate limits as the launch period ends. Users should verify that the model's capabilities, such as robotics simulation, align with their specific workflow requirements before upgrading subscriptions.

Background: Prompt injection attacks

Look for sudden shifts in tone, refusal to follow system constraints, or output that repeats hidden instructions. These signs indicate the model is processing injected commands rather than your intended query.

Read the full guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

What to do and how to stay safe: Google

• Review the vendor's published safety metrics, including prompt injection and red-teaming results, before integrating the model into production environments.
• Monitor API usage costs closely during the launch discount period to prepare for the return to standard pricing rates.
• Verify that the specific tier of subscription you hold includes the features you need, as advanced tools like Antigravity require higher-priced plans.
• Check for updates on guardrail configurations if you are using the model through specialized programs like Fairwind, as access conditions may change.

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

Frequently asked questions

What is the current cost for developers to use Gemini 4 Argon?

Developers pay $2 per million input tokens and $10 per million output tokens, which is a temporary launch discount against standard rates of $4 and $20.

Is Gemini 4 Argon available for free users?

No, consumers need a Pro subscription starting at $19.99 a month, and Google moved free users to its smallest model four days ago.

How does Google claim the model performs on benchmarks?

Google says Argon ranks first on the Vals Index for finance, coding, legal, and tax work, and scores 87 percent on its own ERQA benchmark for robotics.