Gemini 4 Neon model card spotted staged on Vertex: fast reasoning model with 1M context

lyraxana · x · 2026-11-19

A leaked model card for Google's unannounced Gemini 4 Neon has appeared staged on Vertex. The page describes it as a fast reasoning model built on Gemini 4 Hydrogen, bringing quality close to Gemini 4 Carbon for coding, reasoning, and production agents at Flash-tier speed and cost.

Key specs: 1M-token context window, configurable thinking levels (Low/Medium/High default/XHigh), up to 1M output tokens with Long Decode, native Office/OpenDocument file support, and availability via the Gemini platform and Gen AI SDK across Python, Go, JavaScript, Java, and .NET. Unconfirmed leak.

Original post →

More from Models

Models channel →