Google Launches Gemini 4 Argon With 1 Million-Token Output and Limited Cybersecurity Release
Gemini 4 Argon is Google’s latest frontier AI model for complex workflows, software engineering, and cybersecurity.
The model supports up to 1 million output tokens and is being tested on long-running tasks.
Google is initially limiting access to trusted cyber defenders before a wider rollout.
Oct 1, 2026 - Google has launched Gemini 4 Argon, the first model in its Gemini 4 series and the company’s latest flagship AI system. Google says Argon is built for complex, multi-step workflows, with particular capabilities in software engineering and cybersecurity. The model is currently available only to selected cybersecurity organizations and government partners as Google continues testing its safeguards.
Gemini 4 Argon’s Capabilities
Argon is built to handle long-running tasks that require extended reasoning and large amounts of generated output. According to Google, its maximum output limit is 1 million tokens, allowing a single model run to continue across substantially longer workflows than conventional AI responses.
Google has reported strong results on several internal and industry benchmarks. On Vals, a knowledge-work evaluation, Argon scored 68.9%, compared with 63.1% for OpenAI’s GPT-6 Astra and 65.8% for Anthropic’s Claude Fable 5.1.
On Vibe Code Bench, it scored 91.9% ahead of the two competing models cited by Google. The company also acknowledges that Argon does not lead every benchmark, including some tests such as Terminal-Bench 4.0
Google says its own teams are already using Argon for software development, debugging, codebase migrations, and other technical work. The model is also being evaluated for professional workflows involving areas such as legal and financial work.
Focus on Cybersecurity
Cybersecurity is a major part of Argon’s initial rollout. Google says the model can identify, validate, and patch critical software vulnerabilities autonomously, giving security researchers another way to investigate and address vulnerabilities.
Wiz, which Google acquired in 2025, is testing Argon through its Scan for Good initiative. Google said an early evaluation identified a critical vulnerability affecting healthcare software that previous frontier models had not detected.
Availability and Pricing
Google is initially distributing Argon through its Fairwind Program, which provides access to selected cyber defenders, along with the U.S. government. The company is also participating in a U.S. government process for evaluating AI models before broader release.
Broader access is expected to begin with paid API customers and Google AI Ultra subscribers after feedback from early testers. The introductory API price is $2 per million input tokens and $10 per million output tokens. Google is expected to raise those prices to $4 and $20, respectively, after the introductory period.
Google CEO Sundar Pichai said the company plans to expand access as quickly as it can while continuing to strengthen safety measures.
Sources:
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
