Google Unveils Gemini 4 Argon for Coding, Legal Work and Cyber Defense

Google introduced Gemini 4 Argon on Wednesday, September 30, a frontier model aimed at three kinds of enterprise work: large software projects, legal and financial knowledge work, and cyber defense. It arrives at an introductory price that matches the mid-tier models Anthropic and OpenAI shipped in the same week, though Google says the price will later double. For CIOs weighing which models to standardize on, Argon adds a serious option, but most enterprises cannot use it yet.

What is Gemini 4 Argon?

In a blog post, Koray Kavukcuoglu, SVP of Google DeepMind and Google's Chief AI Architect, described Argon as a model for long, multistep workflows. Google lists three target areas: real-world software engineering such as debugging and large codebase migrations, enterprise knowledge work such as financial research and legal drafting, and autonomous cyber defense, including finding, validating and patching vulnerabilities.

Google reports 77.9% on DeepSWE v1.1, a software engineering benchmark, which it calls a new state of the art. It also reports 51.3% on Zapier's AutomationBench, which it ranks first, and a top score of 68% on CWE-bench v1, a vulnerability remediation test, where it says Argon is tied for first. Google says Argon also leads several finance and legal benchmarks, including Harvey's Legal Agent Benchmark, but the post does not give figures for those. All of these are Google's own reported results.

The post says Argon's output limit rises to 1 million tokens, up from 64,000, which matters for tasks like rewriting large code files or drafting long documents in one pass.

How much does Argon cost?

Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper, according to Google. After the introductory period, the price rises to $4 and $20. The post does not say when the introductory period ends.

That launch price lands in a crowded tier. Anthropic released Claude Sonnet 5.5 on September 28 at the same $2 and $10 list price, and OpenAI said at DevDay on September 29 that GPT-6.1 Sol costs one-fifth of GPT-6 Astra's standard token prices, TechCrunch reported. Enterprises building budgets around Argon should plan for the higher rate, not the promotional one.

Who can use Argon now?

Not most companies. Google says Argon is first rolling out to trusted cyber defenders through its Fairwind Program, and that Google is taking part in the US government's voluntary pre-release access process. A broader release will start with paid API customers and Google AI Ultra subscribers, which the company describes as coming soon. The post does not mention Vertex AI, Gemini Enterprise or regional availability.

Google also describes several safety measures. Argon will go to trusted defenders and internal teams without cyber guardrails, while the general release will include misuse safeguards that refuse harmful cyber and chemical, biological, radiological and nuclear requests. Google calls Argon its most prompt-injection-resistant model so far, and says monitoring of the model's reasoning and actions can halt execution. Google adds that the sandboxed environments it uses to train and evaluate models are being isolated and sealed.

What should enterprise leaders do?

Argon's target list, coding, legal, finance and security, is where many large companies already run AI pilots. But a better model does not fix the problems that usually stall those pilots. Ben Sanders of Teradata said at The AI Enterprise Conference 2026 that "the ROI from AI doesn't come from the models," arguing that the gap is business context agents can use.

Practical steps while Argon rolls out:

• Keep model choice flexible. Three major vendors shipped competitive models within three days. Joe McCarron of Skyflow warned in a AI Connect webinar, "You don't want their SLAs to be your SLA," and argued for model-agnostic AI architecture that lets teams switch providers.
• Budget at the full price. If Argon wins your evaluations, model the $4 and $20 rates before committing workloads.
• Test on your own work. Vendor benchmarks are a starting point. A small set of real tasks, scored by domain experts, is a better guide, as HumanSignal's Micaela Kaplan explained in her talk on custom AI benchmarks.
• Ask security teams about Fairwind. Organizations doing vulnerability research may qualify for early access ahead of general release.

The next thing to watch is when Argon reaches Google Cloud's enterprise platforms and how long the introductory price lasts. Both will decide whether it becomes a default model or a specialist tool.

Frequently asked questions

What is Gemini 4 Argon?

Gemini 4 Argon is a Google DeepMind frontier model announced on September 30, 2026, for long, multistep work in software engineering, legal and financial knowledge work, and cyber defense.

How much does Gemini 4 Argon cost?

Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Google has not said when that period ends.

Can enterprises use Gemini 4 Argon now?

Not broadly. Google is rolling it out first to trusted cyber defenders through its Fairwind Program, followed by paid API customers and Google AI Ultra subscribers.

Filed under: AI Strategy & ROI, Security & Fraud, AI Use Cases & Applications