TLDR Tech / Practical
Google's Gemini 4 Argon now supports 1 million token long context
Google has introduced Long Decode for its Gemini 4 Argon model, which enables generating responses up to 1 million tokens by dividing the process across multiple model inferences. This capability utilizes encrypted continuation tokens to preserve the reasoning state and tool usage across different stages. Developers can integrate Long Decode automatically through updated Google Gen AI SDKs or manually via the Gemini API.
Read the full story