Google has introduced Gemini 4 Argon, a new frontier AI model designed for complex, long-horizon workflows spanning software engineering, enterprise knowledge work and cybersecurity defense. Announced on September 30, 2026, Argon expands Gemini’s output capacity from 64,000 tokens to 1 million tokens and is initially being made available to a limited group of trusted cyber defenders through Google’s Fairwind Program.
Google says the phased release is intended to give early testers time to evaluate the model and help improve its safety measures before wider access. The company plans to begin broader availability with paid API customers and Google AI Ultra subscribers, followed by additional developers, enterprises and consumers.
What Google Announced With Gemini 4 Argon?
Gemini 4 Argon is positioned as a model for tasks that require sustained reasoning across multiple steps rather than short, isolated interactions. Google says it is capable across coding, reasoning and multimodal workloads, with applications ranging from debugging and large-scale software migrations to financial research, legal work and cybersecurity defense.
The biggest headline change is its output capacity. Google says Argon can generate up to 1 million tokens in a single output, compared with the previous 64K-token limit. The company argues that the larger output allowance gives the model more room to work through complex problems and produce longer reasoning trajectories without having to split the work into multiple stages.
The distinction is important because the 1 million-token figure refers specifically to the model’s output limit, rather than simply describing a 1 million-token context window.
Gemini 4 Argon Targets Long-Horizon Software Engineering
Software engineering is one of the primary areas Google highlights for Gemini 4 Argon.
Google says its engineers are already using the model for debugging, algorithm design, codebase migration and optimization. Argon agents are also being used for large C and C++ to Rust migration projects, including work involving Google’s Fuchsia Zircon kernel. Google says some of these migrations cover codebases extending to more than 800,000 lines and are subject to automated and manual auditing, emulation testing and review before production deployment.
One example cited by Google involves libgav1, its open-source video decoder. According to the company, Argon agents replaced 32,000 lines of SIMD code in an existing Rust port through repeated profile-guided experiments. Google says the resulting decoder maintained identical video output while running 2.7 times faster than the Rust port.
Google also reports that Argon achieved 77.9% on DeepSWE v1.1, a benchmark focused on real-world, long-horizon software engineering tasks. That figure is a company-reported benchmark result rather than an independently verified industry ranking.
Enterprise Workflows and Multimodal Capabilities
Gemini 4 Argon is not limited to coding. Google says the model is designed for enterprise knowledge work in areas including finance and legal services.
The company reports leading performance on the Vals Index, which evaluates economic-impact-oriented work across finance, coding, legal and tax tasks. Google also cites performance on Vals Finance Agent v2 and Harvey’s Legal Agent Benchmark.
Argon is also designed for workflows involving visual information. Google says the model can analyze professional charts, extract information from long videos and act on information distributed across multiple documents.
On LVBench, a long-video understanding evaluation, Google reports a score of 91.7%. It also reports a 51.3% score on AutomationBench, a benchmark from Zapier that evaluates end-to-end execution across business functions. Again, these figures should be understood as results reported in Google’s launch announcement.
Gemini 4 Argon Adds a Major Cybersecurity Focus
Cybersecurity is one of the most distinctive parts of the Gemini 4 Argon launch.
Google says the model has been trained to autonomously find, validate and patch critical software vulnerabilities. The company is initially providing the model to trusted cyber defenders through its Fairwind Program and says its internal teams will have access to Argon without cyber guardrails so they can use its full defensive capabilities.
Google also says cybersecurity company Wiz is using Argon through its Scan for Good initiative, which focuses on finding and remediating high-risk exposures in critical public infrastructure.
In one early demonstration described by Google, Argon identified a critical vulnerability that exposed sensitive personal information in healthcare software used by hospitals worldwide. Google says the vulnerability had not been identified by previous frontier models.
On CWE-bench v1, Google reports that Argon tied for first place with a score of 68% on vulnerability remediation. The company also says the model outperformed its 3.8 Flash Cyber predecessor on internal vulnerability-discovery and black-box penetration-testing evaluations.
Why Google Is Limiting Initial Access?
The restricted rollout is directly connected to the model’s cybersecurity capabilities.
Google says it is strengthening safeguards against misuse involving cyberattacks and chemical, biological, radiological and nuclear risks. It is also working on defenses against indirect prompt injection attacks, which can attempt to manipulate an AI system through malicious instructions embedded in external content.
Another safeguard involves monitoring the model’s behavior for signs of misalignment. Google says its systems can monitor Argon’s internal reasoning-related signals and actions and stop execution when necessary. The company also says it is hardening sandboxed environments used for high-risk training and evaluations.
Google is participating in the U.S. government’s voluntary process for pre-release model access while it gathers feedback from early testers and continues refining its safeguards.
Gemini 4 Argon Pricing and Availability
Argon is launching with an introductory API price of $2 per million input tokens and $10 per million output tokens. Google says cached input tokens receive a 95% discount from the standard input-token price. After the introductory period, pricing will increase to $4 per million input tokens and $20 per million output tokens.
The model is not yet broadly available. Its initial rollout is focused on trusted cyber defenders and testers through Fairwind. Google says the next stage will begin with paid API customers and Google AI Ultra subscribers before broader availability to developers, enterprises and consumers.
That staged approach makes Gemini 4 Argon different from a conventional model launch where developers and consumers receive access immediately. For now, Google’s announcement combines a major capability expansion with a deliberately controlled deployment strategy.
The central technical change is the jump from a 64K to a 1 million-token output limit, but Google’s broader positioning for Gemini 4 Argon is around sustained work: software engineering, enterprise analysis, multimodal knowledge tasks and autonomous cybersecurity defense. The model’s reported benchmark results and internal deployments provide Google’s evidence for those capabilities, while wider access will depend on the continued testing and refinement of its safety systems.
FAQs
1. What is Gemini 4 Argon?
Gemini 4 Argon is Google’s new frontier AI model for complex, long-horizon tasks involving software engineering, enterprise knowledge work, multimodal reasoning and cybersecurity defense.
2. What is Gemini 4 Argon’s output limit?
Google says Gemini 4 Argon supports an industry-leading 1 million-token output limit, up from 64,000 tokens previously.
3. Is Gemini 4 Argon publicly available?
Not yet in broad form. Google is initially rolling it out to trusted cyber defenders through the Fairwind Program and plans to expand access in stages.
4. How much does Gemini 4 Argon cost?
The introductory API price is $2 per million input tokens and $10 per million output tokens. After the introductory period, Google says pricing will become $4 per million input tokens and $20 per million output tokens.
5. Why is Google restricting the initial rollout?
Google says the phased rollout allows it to gather real-world feedback and strengthen safeguards because of Argon’s advanced cybersecurity and other frontier capabilities.
Also Read –
Gemini 3 Pro: Google’s Most Advanced AI Model Explained
Gemini for Business NotebookLM Integration Explained
Source
Our next era of frontier intelligence Gemini 4 Argon
Gemini 4 Argon official announcement
