Skip to main content

Google's Argon AI Opens to All, Priced at $2 per Million Tokens

Google's Argon AI Opens to All, Priced at $2 per Million Tokens

Google announced on Tuesday that its flagship AI model, Argon, is now available to developers and enterprise customers worldwide. Previously, it was offered only to a select group of trusted cybersecurity professionals through Google's Fairwind program. This phased rollout balances security assessments with commercial promotion, highlighting the importance of risk control when expanding access to cutting-edge models.

Image

Currently, Argon is available to paying users through the Google AI Studio API, Google AI Ultra and Pro subscription services in the Gemini App, and the intelligent agent development platform Antigravity. Its pricing is $2 per million input tokens and $10 per million output tokens, with prices doubling starting in December.

Even after the price increase, it remains cheaper than OpenAI's GPT-6Astra, which costs $10 per million input tokens and $50 per million output tokens. However, according to reports, some Google employees still have doubts about Argon's performance in areas such as coding.

Koray Kavukcuoglu, head of Google DeepMind, said that internal software engineers have used Argon to deliver high-quality code more quickly, and the model has also been used to optimize the user experience of Chrome. Google hopes to demonstrate the value of the model in real engineering tasks. Kavukcuoglu also said that Argon, combining multimodal reasoning, agent coding, and physical understanding capabilities, has the potential to advance embodied robotics research.

In one demonstration, researchers reconstructed the scene layout and motion trajectory from videos of humans throwing sandbags and passed the computational results to a humanoid robot, which then hit the target seven times consecutively in a real environment.

Key Points

  • Global availability: Argon is now open to developers and enterprises worldwide after a limited release to cybersecurity experts.
  • Pricing: $2 per million input tokens and $10 per million output tokens, doubling in December.
  • Competitive edge: Still cheaper than OpenAI's GPT-6Astra, though internal doubts about coding performance persist.
  • Real-world applications: Used internally for code generation and Chrome optimization; potential in robotics.
  • Robotics demo: Humanoid robot hit a target seven times in a row after learning from video.