Google has introduced Gemini 4 Argon, a new frontier AI model focused on cybersecurity, coding, and complex reasoning, currently available to a limited group of testers.
Key facts
- •Gemini 4 Argon is Google's first frontier model release in over seven months.
- •Internal use of the model helped Google save 300 TiB of memory across its data centers.
- •The model has been used to migrate over 800,000 lines of code in the Fuchsia OS Zircon kernel.
- •Google claims Argon scored 77.9 percent on the DeepSWE v1.1 software engineering benchmark.
- •The model supports a 1-million-token output limit, intended to allow for more thorough problem-solving.
Google has announced Gemini 4 Argon, its latest frontier AI model designed for complex tasks including software engineering, legal and financial analysis, and cybersecurity defense. The model is currently being rolled out to a select group of cyber partners through Google’s Fairwind Program. Google reports that its internal teams are already utilizing the model for large-scale codebase migrations and data center optimization.
By the numbers
Capabilities and Performance
Google claims that Gemini 4 Argon outperforms competing models from OpenAI and Anthropic across several AI benchmarks, including the DeepSWE v1.1 software engineering test and the Vals Index for economic analysis. The model is capable of parsing various inputs, including text, images, and long-form video, and is designed to sustain reasoning across complex, long-horizon workflows.
Rollout and Safety
Access to the model is currently restricted to trusted cyber defenders and internal Google teams as part of a phased release strategy. Google is participating in the U.S. government’s voluntary pre-release model access program and plans to implement additional safety measures, such as defenses against prompt injection and misuse, before a broader public release. A specific date for general availability has not been announced.
Pricing and Technical Specifications
Google has set introductory API pricing at $2 per million input tokens and $10 per million output tokens, with a 95 percent discount for cached input tokens. The model supports an output limit of 1 million tokens, a significant increase from the 64,000-token limit in previous versions. To support these long-form outputs, Google is introducing a 'Long Decode Continuation' feature to manage response pauses and resumes.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by TechCrunch, Ars Technica, The Verge, The Decoder.
