Skip to main content

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

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

Google announced Tuesday that Argon, its flagship AI model, is now available to developers and enterprise customers around the globe. Until now, access was limited to a hand-picked group of cybersecurity pros through Google's Fairwind program. The wider release marks a careful balance between security testing and commercial rollout—a sign that Google wants to keep a tight grip on risk as it opens the gates to cutting-edge tech.

Where You Can Get It and What It Costs

Paying users can tap Argon through the Google AI Studio API, Google AI Ultra and Pro subscriptions in the Gemini App, and the Antigravity agent development platform. The sticker price: $2 per million input tokens and $10 per million output tokens. Come December, those rates double.

Even after the hike, Argon undercuts OpenAI's GPT-6 Astra, which charges $10 per million input tokens and $50 per million output tokens. But price isn't everything. Reports suggest some Google employees still have reservations about how well Argon handles tasks like coding.

DeepMind Chief Touts Real-World Wins

Koray Kavukcuoglu, head of Google DeepMind, begs to differ. He says internal software engineers have used Argon to ship high-quality code faster, and the model has even been put to work optimizing Chrome's user experience. Google wants to show the model's value in gritty engineering tasks, not just benchmarks.

Kavukcuoglu also points to Argon's blend of multimodal reasoning, agent coding, and physical understanding as a potential springboard for embodied robotics research. In one demo, researchers reconstructed the scene layout and motion trajectory from videos of humans throwing sandbags, then fed the computational results to a humanoid robot. The robot went on to hit the target seven times in a row in a real environment.

What This Means for Developers

For developers, Argon's global availability opens a new option—one that's priced aggressively against competitors. But the looming price increase in December means teams will need to weigh costs carefully. And the internal doubts about coding performance suggest Argon may not be a slam dunk for every use case.

Key Points:

  • Google's Argon model is now available worldwide to developers and enterprises.
  • Pricing starts at $2 per million input tokens, doubling in December.
  • It's cheaper than OpenAI's GPT-6 Astra but faces internal skepticism over coding.
  • DeepMind highlights real-world applications, including code generation and robotics.
  • A humanoid robot demo showed Argon's potential in embodied AI.