Google’s Gemini 4 Argon Targets Cyber Defense, Coding, and Enterprise AI Work

Google announced Gemini 4 Argon, the first model in its Gemini 4 series, calling it the company’s “next era of frontier intelligence.” The model is aimed at high-stakes work including coding, cybersecurity defense, and enterprise knowledge work in areas such as legal and finance. The launch marks Google’s latest effort to reassert itself in the race for frontier AI systems.

Business Insider reports that Google has fallen behind rival models from OpenAI and Anthropic in areas including engineering and cybersecurity, despite updates to its Flash model series. Gemini 4 Argon is positioned as a potential step back to the frontier. The context is notable because Google previously planned a Gemini 3.5 Pro release, but that model was postponed internally and ultimately not released.

Google is initially making Argon available to trusted partners through its Fairwind Program, which involves vetted governments and cyber authorities testing new models and addressing cyber vulnerabilities. The company said Argon ties with OpenAI’s GPT-6 Astra for the highest score on CWE-bench, a test focused on finding and patching security vulnerabilities. Google also said the model set a new high score on a test of real-world, long-horizon engineering tasks.

Google did not say when Gemini 4 Argon will be publicly available. The company said it is “actively engaged” in the US government’s early-access framework for assessing cybersecurity and other frontier-model risks before launch. Internally, Google said employees are using Argon for debugging and large-scale codebase migrations, and that it helped engineers free up more than 300 tebibytes of memory across data centers without new hardware. Google also said it added systems to monitor Argon’s chain of thought, stop it when necessary, and avoid feedback methods that could teach it to evade monitoring.