Overview

Google introduced Gemini 4 Argon, a new frontier AI model, on September 30, 2026. The model is designed to handle long, complex workflows that require reasoning across multiple steps.

According to Google, Gemini 4 Argon is built for professional workflows including software engineering, finance, legal knowledge work, and cybersecurity defense. Its rollout is currently in an early, controlled stage.

Gemini 4 Argon

A New Frontier Model

Gemini 4 Argon is a new Gemini model developed by Google DeepMind. Instead of focusing only on individual questions, it is designed to work through longer tasks that can involve multiple steps.

Google says its engineers are already using Argon for tasks such as debugging, code migration, and algorithm development.

1 Million Token Output Limit

One of Gemini 4 Argon's notable technical features is its 1 million token output limit. Google says this is an increase from the previous 64K-token limit.

The larger output capacity is designed to support long documents, complex reasoning, and extended multi-step workflows.

Coding and Professional Work

Software Engineering

Gemini 4 Argon is designed for software engineering workflows. Google says its engineers have used the model for debugging, large-scale code migration, and algorithm development.

Google also gives examples involving the migration of C/C++ codebases to Rust.

Finance and Legal Work

The model is also designed for professional knowledge work in areas such as finance and law.

Google's examples include financial research, legal research, and drafting tasks that can involve multiple steps.

Visual Understanding

Gemini 4 Argon is a multimodal model that can work with visual information in addition to text.

Google's examples include analyzing professional charts, finding information in long videos, and combining information from multiple documents to complete tasks.

Cybersecurity Capabilities

Vulnerability Detection

Gemini 4 Argon has also been developed for cybersecurity defense. Google says the model can identify software vulnerabilities, validate them, and create patches for some critical vulnerabilities.

These capabilities are being used in controlled environments involving trusted cyber defenders and Google's internal teams.

Security Testing

Google has evaluated Argon through cybersecurity benchmarks and internal testing.

The company says it is continuing to strengthen security safeguards before expanding access more broadly.

Safety and Security

Controlled Rollout

Gemini 4 Argon is not currently a general public release. Google is initially making it available to trusted cyber defenders and selected testers.

The phased approach allows Google to collect real-world feedback and continue evaluating the model's safety.

Prompt Injection Protection

Google says Argon has been improved to provide greater robustness against indirect prompt injection attacks.

The company also describes the use of red-team testing and adversarial training as part of its security work.

Misalignment Monitoring

Google says monitoring and mitigation systems are used to help keep the model's actions aligned with the user's intended task.

The company also describes mechanisms that can stop execution when necessary.

Performance Information

Google-Reported Benchmarks

Google reports that Gemini 4 Argon achieved a 77.9% score on DeepSWE v1.1, 51.3% on AutomationBench, and 91.7% on LVBench.

These are benchmark results reported by Google. Because each benchmark measures different capabilities, the numbers should not be treated as a single overall performance score.

Cybersecurity Benchmark

Google reports a 68% result for Argon on CWE-bench v1, a benchmark focused on software security vulnerability remediation.

Availability and Pricing

Current Availability

Gemini 4 Argon is currently being rolled out to trusted cyber defenders and selected testers.

Google says availability will later expand to developers, enterprises, and consumers. The company has also stated that initial public availability will begin with paid API customers and Google AI Ultra subscribers.

API Pricing

Google announced the following introductory API pricing:

• Input: $2 per 1 million tokens
• Output: $10 per 1 million tokens
• Cached input: 95% discount from the input-token price

After the introductory period, Google says pricing will be $4 per 1 million input tokens and $20 per 1 million output tokens.

Google's Internal Use

Google says Gemini 4 Argon is already being used in internal workflows.

Examples provided by Google include quantum algorithm optimization, data-center memory optimization, and large-scale software code migration.

Final

Gemini 4 Argon is Google DeepMind's new frontier AI model, with a focus on long reasoning workflows, software engineering, professional knowledge work, multimodal understanding, and cybersecurity defense.

The model is currently in a controlled rollout rather than a full public release. Google plans to expand access to developers, enterprises, and consumers as the rollout progresses.