Google Opens Gemini Argon 4 to Global Developers at $2 Per Million Tokens

Global availability for Google’s latest flagship AI model, Gemini Argon 4, was confirmed on Tuesday. The release follows a restricted testing phase where access was limited to a select group of "trusted cyber defenders" through the company’s Fairwind Program. This staged rollout aligns with current industry practices among frontier AI laboratories, which often balance safety protocols with promotional strategies amid growing concerns about misaligned artificial intelligence.

Access Tiers and Pricing Structure

Developers and business customers can now integrate Argon via the API in AI Studio, while subscribers to Google AI Ultra and Pro gain access through the Gemini App. The model is also available on Antigravity, Google’s agentic software development platform. Current pricing stands at $2 per million input tokens and $10 per million output tokens. However, these costs are scheduled to double starting in December.

Even with the upcoming price increase, Argon remains significantly less expensive than competing frontier models. For comparison, OpenAI’s GPT-6 Astra charges $10 per million input tokens and $50 per million output tokens. This cost differential may provide Argon with a competitive advantage, although its launch has been tempered by reports of internal employee skepticism regarding the model’s performance in specific areas, including coding tasks.

Internal Adoption and Technical Claims

Google DeepMind lead Koray Kavukcuoglu highlighted extensive internal usage in the announcement, stating that teams have applied Argon to complex engineering challenges. "Over the past few weeks, software engineers have been shipping high-quality code faster than ever," Kavukcuoglu wrote. He noted that the model has also contributed to streamlining the Chrome user experience.

Kavukcuoglu described Argon as combining multimodal reasoning, agentic coding, and "deep physical understanding." These features are positioned to drive advancements in embodied robotics, a sector some experts view as essential for achieving true artificial general intelligence. The theory suggests that only agents capable of direct environmental interaction can develop sufficiently dynamic world models to adapt to novel challenges.

Robotics Demonstration and Market Context

In a practical demonstration, Argon processed video footage of a human playing cornhole, mapping the room’s layout and the player’s motor movements. These calculations were transferred to a humanoid robot, which Kavukcuoglu reported "quickly sank seven consecutive shots in a real-world trial."

The broader context of this release includes previous setbacks for Google’s AI division. While competitors such as OpenAI, Anthropic, and Meta have consistently launched new products recently, Google’s Gemini 3.5 Pro faced delays as the company worked to enhance its capabilities. Additionally, Google has experienced departures among senior researchers, including longtime chief scientist Jeff Dean.

For most users, however, the primary utility of AI lies in everyday tasks rather than robotics. To achieve mainstream adoption, Google must demonstrate that Argon offers value comparable to existing tools like Claude, ChatGPT, and Muse, which millions already use for routine activities. The introduction of a free tier could further support this goal.