Google Unveils Gemini 4 Argon but Keeps It Behind Closed Doors
Google has officially launched Gemini 4 Argon, a new AI model focused on cybersecurity, coding, research, and advanced reasoning tasks. The company says Argon was specifically trained for defensive cyber operations and is capable of autonomously identifying, validating, and patching critical software vulnerabilities. Access to the model is currently restricted to trusted cybersecurity partners through Google's Fairwind Program. Beyond security applications, Google reports that Argon is already being used internally for debugging, engineering workflows, and large-scale code migrations. The model also features strong multimodal capabilities, including analyzing long videos, charts, and other visual content. Google claims Argon outperforms competing models from OpenAI and Anthropic across multiple benchmarks and currently leads the Vals AI model rankings. The launch comes as competition among major AI labs intensifies, with firms racing to release increasingly capable models despite ongoing concerns about safety and misuse. Argon's debut also highlights Google's growing momentum in AI, following the announcement that Gemini has surpassed one billion monthly users.