Google unveiled Gemini 4 Argon for complex, long-horizon workflows.
Google says Argon is available to trusted cyber defenders and testers through Fairwind.
Google says Argon will roll out in phases as it gathers feedback on guardrails.
Google expects introductory pricing of $2 per million input tokens and $10 per million output tokens.
Google says Argon's output limit is 1 million tokens, up from 64,000.
Bloomberg reports some internal engineers found Argon weaker in practice than on benchmarks.
Google has unveiled its long-anticipated model, Gemini 4 Argon, which is now available to a group of "trusted cyber defenders" and testers through its Fairwind Program.
In a blog post published Wednesday, Koray Kavukcuoglu, SVP of Google DeepMind and Google's chief AI architect, said that the model is designed to sustain deep reasoning across complex, long-horizon workflows.
"It delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense," said Kavukcuoglu.
The new model will be rolled out in phases. The team said it will continue to gather feedback from early testers to iterate on guardrails.
"Safely releasing frontier capabilities at this level requires a phased approach," said Kavukcuoglu. "We are actively engaged in the U.S. government’s voluntary process for pre-release model access while we gradually expand access."
The Gemini 4 Argon model is expected to launch at an introductory price of $2 per million input tokens and $10 per million output tokens, according to the post. Cached input tokens will be priced at 95% off the input token price.
Enhanced capability
The new model comes with an expanded output limit of 1 million tokens, up from 64,000 previously, according to the post.
"When the model has the headroom to think deeply and generate hundreds of thousands of tokens in a single trajectory, it adds a new level of depth in reasoning to solve tough problems in one go," said Kavukcuoglu.
Kavukcuoglu noted that Google engineers have been using Argon for their daily tasks, including debugging, codebase migrations, and algorithm designs.
However, Bloomberg reported Wednesday that some internal engineers were skeptical, saying Gemini 4 performed well on benchmarks used to gauge model efficacy but fared less well when staff actually put it to work.
Google also highlighted the new model's visual understanding capabilities. It can drive chart analysis, identify video details, and take action based on a series of documents, according to Kavukcuoglu.
The model is also "highly capable at cybersecurity defense," and can autonomously identify, validate, and patch software vulnerabilities, said Kavukcuoglu.
For example, security firm Wiz is already using Argon for cybersecurity defense. Google said the model uncovered a vulnerability exposing sensitive personal data across healthcare software used by hospitals worldwide.
The team also said the model is designed to "refuse harmful requests" to prevent bad actors from using it for attacks.
"These safeguards underwent robustness testing by internal and external red teams using a combination of manual and automated attack methods," Kavukcuoglu added.
