Google Launches Gemini 4 Argon With a 1 Million Token Output Limit, Starting With Cyber Defenders

Arthur Kay
AI

Google DeepMind announced Gemini 4 Argon on September 30, a new frontier AI model built for long, multi-step tasks in software engineering, defensive cybersecurity, and enterprise knowledge work.

According to Google, Argon is rolling out first to vetted cyber defenders through the company's Fairwind Program, with paid API tiers for developers and Google AI Ultra subscribers next in line.

Gemini 4 Argon is not currently available to free Gemini users, and Google has not given dates for the later rollout stages.

Gemini 4 Argon Raises the Output Limit to 1 Million Tokens

The main technical change is the output ceiling. Gemini 4 Argon can generate up to 1 million tokens in a single run, compared with the 64,000 token limit of previous Gemini generations.

Output capacity sets how much a model can write and reason through in one response; smaller limits force it to shorten answers or split large jobs into fragments.

Google positions Argon for long horizon work where the model has to keep context across a long chain of steps, such as auditing an entire codebase or carrying a migration project through to completion, rather than answering single prompts quickly.

How Google Is Already Using Argon Internally

Google says it has been running Argon across its own engineering and infrastructure teams. The company lists these deployments:

  • migrating legacy C and C++ code to Rust, including core media libraries and more than 800,000 lines of the Fuchsia Zircon kernel
  • optimizing the Rust version of libgav1, Google's open source video decoder, which the company claims now runs 2.7 times faster with identical output
  • analyzing server fleet telemetry, where teams of Argon agents found memory optimizations that freed more than 300 terabytes, with projected savings of up to 1 petabyte

On the security side, Google says Argon found a severe vulnerability exposing sensitive data in hospital software used worldwide, while working with security firm Wiz through its Scan for Good initiative. Earlier frontier models missed the flaw entirely, according to the company.

The phased rollout is tied to the compute requirements and strict safeguards that a model with this level of autonomy needs. Cyber defenders in the Fairwind Program get access before the wider developer market does.

Developers who want to build on Argon will need a paid Gemini API tier once access opens there. Consumers will need a Google AI Ultra subscription, as cheaper Gemini plans are not part of the announced rollout.

The launch follows OpenAI's GPT-6 Sol and Luna release in September. Google has not published API pricing for Gemini 4 Argon or a date for availability in Google AI Ultra.

Comments

  1. daniel said on
    Reply

    Raising the output ceiling to 1M tokens only helps if the model keeps a consistent plan across all those steps; longer output without reliable multi-step reasoning just gives you more text to audit. It will be interesting to see independent tests on real codebase audits instead of Google’s internal rollout. For anyone tracking how these model limits apply to practical creative work, I have been following the tool comparisons at migos-aivideo.com.

Leave a Reply

By commenting, you accept our privacy policy and keep the discussion respectful and on topic.