2 min read
Add as a preferred source on Google

Google's Gemini 4 Argon Processes Nearly 25 Trillion Tokens in 72 Hours

The total is roughly three times Gemini 3.0's comparable volume. Google says it faces capacity constraints while serving the model.

Unmarked ceramic tiles spill through a narrow metal chute. / TokenPost.ai
Unmarked ceramic tiles spill through a narrow metal chute. / TokenPost.ai

Google's Gemini 4 Argon processed nearly 25 trillion tokens in its first 72 hours, bringing the company's computing capacity into focus as it expands access to the AI model.

That volume was roughly three times what Gemini 3.0 processed over the same length of time. Tokens are units of input sent to an AI model or output it produces. They are not tradable crypto assets, and the total measures model use rather than revenue.

Alphabet CEO Sundar Pichai said Google is “capacity constrained as we are serving it.” Separately, Google is forgoing some near-term Google Cloud revenue as it directs computing resources toward frontier AI development. That broader allocation was not identified as a cost of serving Argon specifically.

The token figure cannot be converted into API sales using the model's published prices. Argon is available through subscription products as well as a paid API, and the API charges different rates for input and output tokens. The 72-hour total does not break down use by access route or token type.

Google first offered Argon to selected cybersecurity defenders on Sept. 30. It expanded access to developers and businesses on Oct. 13. The model is designed for extended tasks that include coding and research.

In the Gemini app and AI Mode in Search, Argon is available to Google AI Ultra and Pro subscribers. Access through Antigravity is available to Ultra subscribers. Developers can also use the paid API, which bills for tokens processed rather than granting access through those subscription plans.

For the standard public-access Argon model, introductory API rates are $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95% discount during the introductory period. The different rates are another reason the overall token count does not establish how much API revenue the model generated.

The introductory rates run through Nov. 30, 2026. After that date, Google plans to raise the standard public-access rates to $4 per million input tokens and $20 per million output tokens.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…