Google has officially entered the next phase of its Gemini lineup with Gemini 4 Argon, a new frontier AI model designed for complex, long-running tasks. The company says Argon delivers major improvements in software engineering, enterprise knowledge work, research and cybersecurity. It is also built to reason through much longer workflows instead of stopping after a few steps.
The launch comes as competition in the AI industry continues to intensify. Google had faced questions about the pace of its next-generation models, but CEO Sundar Pichai had earlier indicated that the company was working on its next major release. Gemini 4 Argon is now here, although Google is taking a gradual approach to its wider rollout.
Gemini 4 Argon Specifications
| Specification | Gemini 4 Argon |
|---|---|
| Model Family | Gemini 4 |
| Model Name | Gemini 4 Argon |
| Focus Areas | Coding, Knowledge Work, Cybersecurity, Research |
| Output Token Limit | Up to 1 Million Tokens |
| Initial API Price | $2 per 1M Input Tokens |
| Initial Output Price | $10 per 1M Output Tokens |
| Cached Input | 95% Discount |
| Wider Access | Paid API Customers, Google AI Ultra |
| Initial Availability | Trusted Cyber Defenders |
| Security Programme | Google Fairwind |
One of the biggest upgrades is the model’s 1 million-token output limit. Google says this is a major increase from the previous 64,000-token limit. It allows Argon to reason through much longer workflows and potentially generate hundreds of thousands of tokens in a single trajectory.
That capability is aimed at tasks where an AI model needs to maintain context for a long time. This includes large software projects, detailed research, financial analysis and other professional workloads.
Gemini 4 Argon Shows Strong Benchmark Performance
Today we’re introducing Gemini 4 Argon.
It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit. pic.twitter.com/oDaCoOLu5E
— Google (@Google) September 30, 2026Google’s published benchmarks show Argon performing strongly across several categories. On the Vals Index, which evaluates knowledge work across areas such as finance, coding, legal and tax tasks, Argon scores 68.9%. GPT-6 Astra scores 63.1%, while Claude Fable 5.1 scores 65.8%.
Coding is another major focus. On Vibe Code Bench, Gemini 4 Argon scores 91.9%, compared with 90.3% for Claude Fable 5.1 and 89.6% for GPT-6 Astra. Google also reports a 77.9% score on DeepSWE v1.1, a benchmark focused on long-horizon software engineering tasks.
📊 Gemini 4 Argon Benchmark Showdown!#Google’s latest AI model #Gemini4Argon delivers strong results across knowledge work, coding, science, long-context reasoning, multimodal understanding and cybersecurity. 🚀#Gemini4 #GoogleAI #AI #Techotales pic.twitter.com/Yc4yM5bPpG
— Techotales (@techotales) October 1, 2026However, Argon does not lead every benchmark. On Terminal-Bench 4.0, for example, its score is 57.4%, compared with 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5. This shows that performance still varies depending on the type of task and benchmark being used.
Google also reports strong multimodal and long-context performance. Argon scores 91.7% on LVBench for long-video understanding and 84.2% on the GraphWalks test covering contexts between 256K and 1 million tokens.
Gemini 4 Argon Gets Advanced Cybersecurity Skills
Cybersecurity is one of the most important areas for Argon. Google says the model has been trained to help defenders find, validate and patch critical software vulnerabilities autonomously.
Google-owned cybersecurity company Wiz is already using Argon through its Scan for Good initiative. In an early demonstration, Argon reportedly discovered a critical vulnerability exposing sensitive personal information in healthcare software used by hospitals worldwide. Google says previous frontier models had missed the issue.
On CWE-bench v1, which measures vulnerability remediation, Argon records a score of 68%, tying for the highest score in Google’s published comparison.
Limited Release Before Wider Availability
Lots of discussion out there about our next model(!), so I wanted to give an early look as soon as possible. Introducing Gemini 4 Argon!
It shows frontier performance in complex workflows, cyber defense and software engineering. Teams are using it extensively at Google, from… pic.twitter.com/sv4VNmQ0YT
— Sundar Pichai (@sundarpichai) September 30, 2026Google is not opening Gemini 4 Argon to everyone immediately. The model is initially being made available to trusted cybersecurity defenders through the Fairwind Program. Google says the phased rollout is intended to gather real-world feedback and strengthen safeguards before wider availability.
The company plans to make Argon available next to paid API customers and Google AI Ultra subscribers, followed by broader access for developers, enterprises and consumers.
Pricing will initially start at $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95% discount. After the introductory period, the pricing will increase to $4 per million input tokens and $20 per million output tokens.
Conclusion
Gemini 4 Argon marks a significant step in Google’s Gemini strategy. Instead of focusing only on faster answers, Google is targeting complex workflows that can require deeper reasoning, large amounts of context and multiple steps.
Its 1-million-token output limit, coding capabilities and cybersecurity focus make Argon particularly interesting for developers and enterprise users. But the model is still in a controlled rollout, so its performance in everyday real-world use will become clearer as access expands.
For now, Google is positioning Gemini 4 Argon as its new frontier model for the most demanding AI workloads.
Follow Us: Facebook | X | Instagram | YouTube | Pinterest
