After OpenAI introduced new AI agents Dots and changed the system and subscription prices, Google also rushed in with its own news. This year it already introduced Gemini 3.6 Flash or Gemini 3.8 Flash Cyber and now offers users even more advanced Gemini 4. He gave it the unusual nickname Argon and claims it is the most advanced model for programming, knowledge work, cyber defense, and creative writing.
Gemini 4 Argon is a substitute Gemini 3.5 Pro, which was supposed to be released, but Google canceled it. Instead, he opted for a more advanced model and stopped confusing users with decimal numbers. But the most important thing is that it offers a significantly larger number of tokens, so that they can work more intensively, think better, program and write. And it can easily handle even high consumption, because unlike the previous one maxima has 64,000 tokens, making them available equally 1 million.
Better than GPT-6 Astra
But Google won't have it easy with this new feature, because OpenAI introduced its most advanced model at the beginning of September GPT-6 Astra, which has aroused both enthusiasm and horror. It has advanced cyber defense capabilities, which can easily backfire. Moreover, it can conceal its thinking. And the upcoming GPT-6.1 Astra is said to be even more cunning in this regard and tries to hide its thoughts from the user and deceive him. However, Google still dares to claim that Gemini 4 is even more advanced.
You might be interested in
This is proven by published benchmarks, in which Gemini 4 achieves better results in many areas than GPT-6 Astra whether Claude Fable 5.1. "In the CWE-bench v1 benchmark, which evaluates a model's ability to fix security vulnerabilities, Argon tied for first place with the highest score of 68%," says Google. Both competing models are causing concern among experts cyber security great respect.
Only for paying customers
To the strongest points Gemini 4 will include programming, writing, the aforementioned defense, and also understanding long videos. However, we probably all know that it won't be free:
- introductory price is set at 2 dollars per million input tokens and 10 dollars per million output tokens,
- then increases to 4 dollars for 1 million input tokens and 20 dollars for 1 million output tokens.
Gallery
Google will Gemini 4 will be made available gradually, for now exclusively as part of the most expensive subscription Google AI UltraAt the same time, the model will be available to paying API customers. The technology giant claims to have done everything possible to prevent possible abuse, deploying various measures against non-compliance and allegedly using all available means to prevent the model from behaving arbitrarily.
How will the new one be? Gemini It remains to be seen whether it will be a danger to the world and whether it will succeed against Anthropic and OpenAI, but we will find out. So far, it seems to me to be similarly advanced and I do not dare to judge which one will be better. After all, we already know that each company excels in a different area. But it seems to me that Google now wants to improve programming in particular, in which it has so far lagged significantly behind Claude.
