Google has unveiled Gemini 4 Argon, its most advanced artificial intelligence model yet, built for long, complex tasks rather than quick chatbot replies.

The headline upgrade is the model’s output limit: Argon can write up to 1 million tokens in a single response, up from 64,000 tokens in earlier models. Google says the headroom lets the model think and write through hundreds of thousands of tokens in a single run.

The company is positioning Argon for software engineering, cybersecurity and enterprise work. On benchmarks reported by Google, Argon scored 77.9% on DeepSWE v1.1 for long software tasks and 91.7% on LVBench for long-video understanding. These are Google’s own figures and have not been independently verified.

Google says Argon agents are already in use internally, including work on rewriting C and C++ code in the memory-safe Rust language and optimising data-centre memory use.

Argon is not publicly available yet. Google is rolling it out first to selected cybersecurity defenders through its Fairwind Program, with broader access for API customers and Google AI Ultra subscribers planned after further testing.