Google Launches Gemini 4 Argon With Output Up To 1M Tokens

Google unveiled Gemini 4 Argon on September 30. It is a flagship model aimed at development, enterprise work, and cyber defense. The output limit has grown to 1 million tokens, up from 64,000 before. That is nearly a 16x jump. On DeepSWE v1.1, the model set a new record of 77.9% on long development tasks. Access is currently limited to vetted cyber-defense specialists in the Fairwind program. The introductory price is $2 per million input tokens and $10 per million output tokens.

Most developers do not have access yet. That is the main limitation right now. The published benchmarks look interesting, but there is no way to test the model on your own agentic tasks until Google opens up the API. For people building long agent chains, a 1-million-token output limit is genuinely useful. It means fewer interruptions on large jobs, like refactoring a big module in a single pass. Still, these are the vendor's own numbers, not real-world project experience. I will wait for open access before drawing any conclusions about quality compared with what I use every day.