TL;DR
Get tech for your team delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
Google announced Gemini 4 Argon on September 30, 2026, and said it is beginning a phased rollout to trusted cyber defenders through its Fairwind Program. The company reports strong results in coding, enterprise tasks and defensive cybersecurity, but broad availability, independent validation of its benchmarks and several rollout details remain unclear.
Google announced Gemini 4 Argon on September 30, saying the model is beginning a phased rollout to trusted cyber defenders through its Fairwind Program. The company says Argon is designed for long, complex tasks in software engineering, enterprise work and cybersecurity defense; developers, businesses and consumers do not yet have general access.
Google describes Argon as a model built to sustain reasoning across long-horizon workflows. Its output limit is 1 million tokens, up from 64,000 tokens in the previous model, according to the announcement. Google says that capacity can support extended work in a single trajectory, though the announcement does not provide independent evidence about performance across typical customer workloads.
The company reports that Argon scored 77.9% on DeepSWE v1.1, a benchmark for real-world software engineering tasks. Google also says it ranked first on Zapier’s AutomationBench with 51.3%, and scored 91.7% on LVBench for long-video understanding. These are results reported by Google against named evaluations; the announcement does not detail independent replication or how performance may vary in practical deployments.
Google also describes internal engineering projects, including a Rust video decoder it says runs 2.7 times faster than an existing Rust port while producing identical video output. The company says other Argon-assisted projects include code migrations and memory optimizations, but notes that large system rewrites face automated and manual audits, emulation tests and review before production rollout. The planned introductory API prices are $2 per million input tokens and $10 per million output tokens, with cached input tokens priced 95% below the input rate.
A Controlled Start for High-Stakes Tasks
Argon’s announcement combines claims of stronger performance on software and business tasks with an unusually cautious initial release. Google says it is providing access to selected cyber defenders while it gathers feedback and works on safeguards. That approach matters because a model able to locate and validate software vulnerabilities may help security teams find and fix flaws, while also raising questions about how access and misuse risks are managed.
For companies considering AI for coding, legal research or finance, the announcement offers early benchmarks and examples, but not a public product they can evaluate today. The limited rollout means customers must wait for access and assess the model in their own environments. Google’s reported results may guide interest, but benchmark scores alone do not establish reliability, cost or suitability for consequential work.
From Internal Use to Early Access
Google says Argon is already being used internally by thousands of employees for specialized coding, research and writing tasks. Its examples include work on quantum algorithms and migration of C and C++ code to Rust, including efforts involving large codebases. Google says such rewrites are subject to rigorous checks before any production release.
The company frames the Fairwind Program as the first stage of external access, focused on trusted cyber defenders. It also says it is participating in a voluntary U.S. government process for pre-release model access. That process and the staged rollout precede planned access for developers, enterprises and consumers, but Google has not supplied a public launch date.
“Safely releasing frontier capabilities at this level requires a phased approach.”
Access, Safeguards and Results
Google has not said how many testers will receive access, when the API will open to developers, or when enterprise and consumer products will follow. The announcement also does not specify how the model’s cybersecurity capabilities will be bounded outside the trusted-defender program, or describe the safeguards in enough detail to assess them.
The performance figures and engineering outcomes are company-reported claims. Independent evaluations, deployment conditions and comparisons with other models are not provided in the source material. Google also does not explain how the introductory prices will apply across different product plans or whether they may change at general release.
Feedback Before Wider Release
Google says it will collect feedback from early testers and continue refining guardrails while it gradually expands access. The next milestones are a broader rollout to developers, enterprises and consumers, and clearer information about availability and pricing. The company has not announced dates for those steps.
Key Questions
What is Gemini 4 Argon?
Gemini 4 Argon is a Google model announced for long, complex workflows in coding, enterprise knowledge work and defensive cybersecurity. Google says its output limit is one million tokens.
Can the public use Argon now?
No general public availability was announced. Google says the model is beginning a limited rollout to trusted cyber defenders through its Fairwind Program.
What results has Google reported?
Google reports scores including 77.9% on DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on LVBench. These are company-reported benchmark results; independent replication was not included in the announcement.
How much will Argon cost?
Google lists introductory API prices of $2 per million input tokens and $10 per million output tokens. Cached input tokens are priced 95% below the input rate.
When will wider access begin?
Google says it plans to expand access after gathering tester feedback and refining safeguards, but it has not provided a public launch date.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
