Google unveils Gemini 4 Neon — a model for coding and agent tasks
The model features a context window of 1 million tokens and four levels of reasoning depth.

Neon is designed for multi-step workflows: it analyzes context, utilizes tools, and executes sequential actions. The model is available via the Gemini API and Google AI Studio, and has also begun rolling out to business users and subscribers of Google's paid plans.
At the xhigh level, Gemini 4 Neon ranked third in the Artificial Analysis Intelligence Index, trailing larger and slower models. In programming tests, Google reported leading results, including multi-file debugging, refactoring, and extended agent scenarios.
Pricing is set at $3 per million input tokens and $15 per million output tokens. For tokens from the cache, the rate is $0.15 per million. The reasoning levels allow users to balance result quality and token expenditure.
A separate version of Neon for cybersecurity is intended for vulnerability detection, risk assessment, and code issue resolution. Google reported its performance in tests for secure code generation and vulnerability detection. Access to this version is planned to be provided to verified specialists through the Fairwind program.




