Google limits Gemini 4 Argon access to cybersecurity defenders

Google has begun a limited rollout of Gemini 4 Argon, giving trusted cybersecurity defenders access through its Fairwind Program before opening the model to developers, enterprises or consumers. The Gemini 4 Argon cybersecurity access plan is a phased release, with Google saying it is using early testing to refine safeguards for a model it says is designed for extended coding, knowledge-work and cyber-defense tasks.

Google has not said when general customers will be able to use Argon. It said it will collect feedback from early testers and continue work on guardrails while participating in the U.S. government’s voluntary pre-release model-access process. CNBC also reported that the initial launch is limited to select cyber partners.

Who can use Gemini 4 Argon now?

For now, access outside Google is limited to trusted defenders in Fairwind, according to Google. The company has presented the restricted launch as a way to put a model trained for cyber defense in the hands of defenders while building confidence in a broader rollout. Tulsee Doshi, Google’s Gemini model product lead, gave that rationale to CNBC.

The boundary matters because Google is announcing a frontier model without offering ordinary developers or enterprises a release date. Google said it intends to expand availability to those groups and to consumers as soon as possible, but did not provide a timetable or explain the eligibility criteria for Fairwind participants in its announcement.

What is Google changing in Gemini 4 Argon?

Argon has a 1 million-token output limit, compared with the previous 64,000-token limit, Google said. The company says the larger limit is intended to support longer, multi-step workflows in software engineering, legal and financial work, and cybersecurity defense.

Google reported that Argon scored 77.9% on DeepSWE v1.1, an evaluation of real-world, long-horizon software engineering work. Those results and the company’s comparisons with other models are Google-reported claims, not independently validated in the material reviewed for this report.

Google has also set planned introductory API pricing at $2 per million input tokens and $10 per million output tokens. Cached inputs would be discounted by 95%, the company said. It has not established when customers outside the limited rollout can purchase access.

What safeguards is Google testing before broader release?

Google says it is working on protections against misuse, indirect prompt injection, and risks in its sandboxed environments. The company also says it has monitoring intended to watch model reasoning and actions and halt execution when needed. These are company descriptions of its safeguards; the supplied reporting does not independently verify how they perform in deployed use.

Internally, Google says thousands of employees are using Argon, including for data-center memory work and codebase migrations. It said critical rewrites are undergoing automated and manual audits, emulation testing and review before production rollout. That qualification is notable as the company moves a model built for longer autonomous tasks from internal use to selected external defenders.

This story draws on original reporting from SiliconANGLE.