Google launches Gemini 4 Argon, its most capable AI model yet

Google has unveiled Gemini 4 Argon, its newest flagship AI model, built for complex multi-step tasks across software engineering, cybersecurity, and specialized knowledge work.

• Google says Argon scored 77.9% on the DeepSWE v1.1 software engineering test and 68% on the CWE-bench v1 vulnerability benchmark.
• Google said the model achieved 91.7% on the LVBench long-video understanding test and helped migrate over 800,000 lines of code.
• Google acknowledged Argon trails rival models in some benchmarks, including two of four programming tests.
• Google is releasing Argon gradually, first to trusted cybersecurity defenders through its Fairwind program, and has not announced a full public launch date.