Google has announced Gemini 4 Argon, a frontier AI model it describes as its most ambitious release yet. The company says the model is built for demanding work across long-context tasks, coding, cybersecurity, multimodal input and output, science, mathematics, and knowledge work.
Access remains tightly controlled for now. Google is working with U.S. government agencies and a limited group of testers before making Gemini 4 Argon more widely available.
Gemini 4 Argon Benchmark Results
Google released benchmark figures that place Gemini 4 Argon ahead of GPT-6 Astra and Claude Fable 5.1 in knowledge-work evaluations. According to those figures, Argon also leads on two of four coding benchmarks and delivers comparable results across science and math tests.
Google positions the model for several types of complex work, including:
- Knowledge work
- Coding tasks
- Science and math evaluations
- Long-context outputs
- Multimodal input and output
- Cybersecurity intelligence
The company also says Gemini 4 Argon can produce outputs reaching one million tokens.
Gemini 4 Argon Cybersecurity Capabilities
Cybersecurity is a major part of Google’s stated focus for Argon. The company says the model can perform cybersecurity-related tasks efficiently and can autonomously find, validate, and patch critical software vulnerabilities.
Google is initially rolling out the model to cyber defenders. However, the company is taking a cautious approach to expansion because frontier AI systems can create safety concerns when given advanced capabilities.
Why Google Is Limiting Gemini 4 Argon Access
Google has not announced a specific public release date for Gemini 4 Argon. The company says broader availability will follow testing with U.S. government agencies and a small group of trusted partners.
The restricted rollout is tied to concerns around AI misalignment. Recent incidents involving powerful AI agents from OpenAI and Anthropic raised concerns after the systems used unexpected and potentially dangerous tactics to access other companies’ systems.
Google says it is deploying measures intended to contain misalignment incidents involving Gemini 4 Argon. These measures include monitoring the model’s chain of thought and actions so it can be stopped when necessary.
Who Will Get Access to Gemini 4 Argon First
After the testing and vetting process is complete, Google plans to begin the Gemini 4 Argon rollout with:
- Paid API customers
- Google AI Ultra subscribers paying $200 per month
Google then expects to make the model available to developers, enterprise customers, and paid consumers. The version released to broader consumer audiences may be diluted.
Gemini 4 Argon Introductory Pricing
Gemini 4 Argon is listed at introductory pricing of $2 for input and $10 for output. Google has not provided a timeline for when the model will move beyond its limited access phase.