After months of delays, Google is gearing up to get back in the AI game. The company on Wednesday announced its newest frontier model, called Gemini 4 Argon. But you won’t be able to use it yet. Here’s why.

A lot has changed in the 10 months since Google released Gemini 3 — and crushed its competition. One is the introduction of the White House’s voluntary review process for new AI models, with which Google said in a blog post it is “actively engaged.” Another is the increasingly potent cybersecurity risk that new AI models pose for online companies, governments and consumers. Just take a look at all the AI agent hacks we’ve seen this summer.

Right now, Gemini 4 Argon is only available to specific cybersecurity defenders in Google’s Fairwind program. This has become the new way that AI companies roll out new frontier models: first to cybersecurity experts, who use it to shore up their own defenses and report any weaknesses, then to subscribers, and then eventually free users. OpenAI did this with GPT-6 Astra, same as Anthropic with Fable 5.

The slow-roll release is meant to give experts the time to find and patch issues before the AI is let loose on the internet. Past releases gave defenders a few weeks before the model was given a general release. Given that this is Google’s first model to go through this process, it could be a little bit longer. Google did not immediately respond to questions about the time frame of Gemini 4’s release.

Gemini 4: What we know so far

Gemini 4 Argon is the latest AI model from Google, capable of coding, multimodalities (like understanding images and charts) and reasoning. If it follows past releases, we can expect Gemini 4, or a version of it, to become the new model powering Google’s many, many AI features in Search, Docs and more. Google said that Gemini 4 Argon is particularly adept at business-related tasks, like software engineering, cybersecurity and knowledge work in legal and finance fields. Benchmarking data released by Google, comparing Gemini 4 to OpenAI’s Astra and Anthropic’s Fable 5.1, seems to back up those claims.

Google says it has been using Gemini 4 internally to help make the company more efficient, including in its quantum computing research and memory operations. It has a 1-million-token output limit. More tokens mean the AI can do more work and handle more complex tasks.

Google was rumored to release Gemini 3.5 Pro earlier this year, but reports say it held off to keep working on it. Even this week, there were concerns within the company that Gemini 4 wasn’t up to par on key tasks, such as coding, Bloomberg reports. We’ll have to wait until the model is released and put it to the test to see if those concerns were addressed before then.

Katelyn Chedraoui

Reporter 2

Katelyn is a reporter with CNET covering artificial intelligence, including chatbots, image and video generators. Her work explores how new AI technology is infiltrating our lives, shaping the content we consume on social media and affecting the people behind the screens. She graduated from the University of North Carolina at Chapel Hill with a degree in media and journalism. You can reach her at kchedraoui@cnet.com.

See full bio