Google Unveils Gemini 4 Argon AI Model to Select Cybersecurity Partners

Google's latest AI model, Gemini 4 Argon, is being rolled out to selected cybersecurity partners while developers and subscribers wait for access. The company has already used it internally for software work and security tasks.

The model's output ceiling has been raised to 1 million tokens, allowing for longer text generation in one run. Google engineers have reported encouraging results with the new AI, including a 2.7 times speed increase on video-decoding software. However, some employees have expressed concerns about its performance on certain coding tasks.

Google's coding comparison shows Argon leading OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5 on DeepSWE v1.1, but trailing behind on FrontierSWE v2 and Terminal-Bench 4.0.