Google has announced its most capable model to date, positioning it as the company’s answer to the growing demand for AI systems that can handle genuinely difficult, sustained professional tasks rather than simple question-and-answer exchanges.

Gemini 4 Argon arrives with a specific focus on real-world software engineering, enterprise knowledge work, and cybersecurity defense, and its initial rollout is unusually narrow by design.

Google Debuts The Latest Gemini 4 Argon Model

Koray Kavukcuoglu, Senior Vice President of Google DeepMind and Chief AI Architect at Google, announced Gemini 4 Argon on the Google blog, describing it as the company’s next era of frontier intelligence. Rather than launching to the general public immediately, the model is being rolled out first to a select group of trusted cyber defenders through Google’s Fairwind Program, the company’s proactive cybersecurity initiative, before expanding to paid API customers and Google AI Ultra subscribers.

Kavukcuoglu said the model is fundamentally changing how Google’s own teams work and build internally, with thousands of Google employees already using it for specialized coding tasks, deeper research, and higher-quality writing. Argon is launching at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95 percent off the input token price.

After the introductory period, pricing is expected to rise to $4 per million input tokens and $20 per million output tokens.

What is Gemini 4 Argon Being Used For?

Gemini 4 Argon is already being used inside Google across a range of demanding internal tasks. Google’s quantum computing team has used Argon to optimize the spacetime resources of subroutines that were previously bottlenecking important applications, beating a published baseline by 40 percent in minutes.

Argon agents have also analyzed fleet-wide profiling telemetry across Google’s data centers to autonomously identify and apply memory optimizations, freeing up more than 300 terabytes of memory with an estimated 500 terabytes to 1 petabyte in total expected savings.

The model has also been used for large-scale codebase migrations, including work to move Google’s C and C++ codebases to Rust across multiple projects. On one of those efforts involving libgav1, Google’s open-source video decoder, Argon replaced 32,000 lines of SIMD code through profile-guided experiments, producing a memory-safe decoder that runs 2.7 times faster than an existing Rust port while maintaining identical video output.

Gemini 4 Argon Brings Deep Reasoning

One of the defining technical features of Argon is its expanded output token limit. Google raised the ceiling from 64,000 tokens to an industry-leading 1 million tokens, giving the model the room it needs to sustain deep reasoning across complex, multi-step problems without running into the kind of truncation issues that plague shorter context windows.

Google describes this as adding a new level of depth in reasoning, allowing the model to solve difficult problems in a single extended trajectory rather than requiring users to break tasks into multiple sessions.

On the benchmark side, Argon sets a new record on DeepSWE v1.1 with a score of 77.9 percent, which measures performance in real-world long-horizon software engineering tasks. It also leads the Vals Index, which measures economic impact across finance, coding, legal, and tax work, weighted by each sector’s contribution to US GDP.

Additional leading results were recorded on Vals Finance Agent v2, Harvey’s Legal Agent Benchmark for legal research and drafting, and AutomationBench, Zapier’s benchmark for end-to-end execution across core business functions, where Argon scored 51.3 percent to rank first.

Where Is Gemini 4 Argon Debuting?

The initial rollout of Gemini 4 Argon is happening specifically through Google’s Fairwind Program, which is designed to give trusted cybersecurity defenders early access to frontier model capabilities for defensive purposes before broader availability. Argon is being released to Fairwind participants without cybersecurity guardrails, allowing them to leverage its full frontier-level capabilities for cyber defense rather than operating within the restrictions Google applies to general consumer releases.

Wiz, a cybersecurity company, is already using Argon through the Fairwind Program as part of its Scan for Good initiative, which is dedicated to protecting critical public infrastructure for free by identifying and remediating high-risk exposures. In an early demonstration, Argon uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, a risk that previous frontier models had failed to detect.

On CWE-bench v1, which measures a model’s ability to remediate security vulnerabilities, Argon tied for first place with a top score of 68 percent. It also outperformed earlier models on Wiz’s own internal black-box penetration testing benchmark, which tests the model’s ability to analyze live web systems without access to their source code.

Frequently Asked Questions

What Is Google Gemini?

Google Gemini is Google’s family of large language models, developed primarily by Google DeepMind. The Gemini name replaced the earlier Bard branding in 2024, and the model family spans a wide range of sizes and capabilities, from smaller, faster models optimized for everyday tasks and mobile devices to frontier models designed for the most demanding professional and research workloads.

Gemini powers Google’s AI features across products including Google Search, Google Workspace, the Gemini app, and Google Cloud, and is available to developers through the Gemini API.

How Powerful Is Google’s Gemini?

Gemini 4 Argon represents Google’s most capable model to date across the benchmarks the company has published. It outperforms previous Gemini generations and matches or exceeds competitors on evaluations covering real world software engineering, legal and financial knowledge work, cybersecurity vulnerability discovery, and long video understanding, where it scored 91.7 percent on LVBench.

Its 1 million output token limit, which is substantially higher than what most frontier models currently offer, is designed specifically to support the kind of sustained, multi-step reasoning that complex professional workflows require.

Internally, Google describes Argon as being in a different capability tier compared to the models it has released before, one that fundamentally changes how the company’s own engineering and research teams operate.

When Is Gemini 4 Argon Coming to the Public?

Gemini 4 Argon is not available to the general public yet. Google is using a staged rollout, beginning with trusted cyber defenders through the Fairwind Program and then expanding to paid API customers and Google AI Ultra subscribers. The company has said it wants to gather feedback from early testers and iterate on guardrails before making Argon broadly available to developers, enterprises, and consumers.

Google did not specify a firm date for the wider public launch, saying only that the model would become available to more users as soon as possible following the initial testing phase. The company is also participating in the US government’s voluntary process for pre-release model access as part of its broader safety review before the general release.