Google DeepMind has released Gemini 4 Neon, a reasoning model designed for coding agents, science, and knowledge work. Positioned between the Gemini 4 Carbon and Hydrogen models, Neon is intended to provide frontier-level reasoning at the speed and cost of the Flash series.
The model is available to AI Pro and AI Ultra subscribers, as well as via Google-hosted APIs, where it replaces the Argon model and 3.1 Pro. It features a 1 million-token unified context window and four adjustable thinking levels.