Story perspectives

Google Unveils Gemini 4 Argon With New Token Pricing

By Drooid · Oct 2, 2026, 2:43 PM UTC · How we work

Summary

Google Launches Argon

Google launched the Gemini 4 Argon frontier AI model.
API pricing is $2 per million input tokens and $10 per million output tokens.
Argon can generate up to one million output tokens, far above the prior 64,000-token limit.
Argon scored 77.9% on DeepSWE, 51.3% on AutomationBench, 91.7% on LVBench, and tied OpenAI on a cybersecurity test.
Google first offers Argon to cybersecurity defenders via the Fairwind program, then to paid API users.

Summary

Hassabis Steps Down

DeepMind founder Demis Hassabis stepped down as CEO, succeeded by Koray Kavukcuoglu.
Training runs for Gemini models can cost up to $400 million, analysts say.
Google employees doubt Argon’s coding performance, with some saying it lags rivals and others calling it frontier-ready.
Experts warn Google is ‘benchmaxxing’, prioritizing benchmark scores over real-world functionality.
Argon’s large size could strain Google’s margins because of higher operating costs.

History

Historical Context

Google bought London-based DeepMind in January 2014 for about $400 million, giving it the AlphaGo team that defeated Lee Sedol 4-1 in March 2016. That win proved reinforcement-learning AI could master complex strategy, prompting Google to launch its own large-language models. Today the Gemini 4 Argon model, priced at $2 per million tokens, extends that legacy while sparking internal cost-efficiency debates.

1 of 3

Google Launches Argon

Google launched the Gemini 4 Argon frontier AI model.
API pricing is $2 per million input tokens and $10 per million output tokens.
Argon can generate up to one million output tokens, far above the prior 64,000-token limit.
Argon scored 77.9% on DeepSWE, 51.3% on AutomationBench, 91.7% on LVBench, and tied OpenAI on a cybersecurity test.
Google first offers Argon to cybersecurity defenders via the Fairwind program, then to paid API users.

1 / 3