Google Releases Gemini 4 Neon for Complex Engineering and Science

Google released Gemini 4 Neon on Wednesday positioning the paid model between Hydrogen and Carbon for complex software engineering and scientific tasks, according to release notes from the Gemini API.

The stable endpoint gemini-4-neon is available through generateContent and the Interactions API. It also appears in the Google AI Studio Pro and Ultra plans, as well as the Google Antigravity managed agent, according to Unite.

The rollout follows a rapid sequence of model releases that began on October 13, 2026, with Argon. Carbon debuted on November 10, 2026, followed by Hydrogen on November 12, 2026.

Google DeepMind executive Koray Kavukcuoglu explained the division of responsibilities across the architecture. “Carbon for projects difficult and prolonged, Hydrogen for fast execution and Neon as balance between depth and efficiency,” Kavukcuoglu told Google Discovery.

Neon is priced at $3 per million input tokens and $15 per million output tokens, occupying a middle tier in the Gemini 4 family. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, maintaining free API access that Neon lacks.

Google Releases Gemini 4 Neon for Complex Engineering and Science
Photo: Google Discovery

Carbon and Argon currently carry introductory rates of $2 per million input tokens and $10 per million output tokens. Those promotional prices run until November 30, 2026, before doubling to $4 per million input tokens and $20 per million output tokens on December 1, 2026.

Gemini 4 Model Input Token Price (per million) Output Token Price (per million)
Hydrogen $1.50 $7.50
Argon / Carbon (Promotional) US$ 2 US$ 10
Neon US$ 3 US$ 15
Carbon (Standard) US$ 4 US$ 20

Neon costs 25% less than Carbon once Carbon’s promotional pricing expires on November 30, 2026, according to Google Discovery. Implicit caching charges $0.15 per million stored input tokens, applying a 95% discount.

Shared 1,048,576-Token Window Caps Synchronous Replies at 65,536

Input, reasoning, and output tokens share a single 1,048,576-token window. Synchronous responses are restricted by default to 65,536 tokens, which counts both reasoning and final output tokens toward the limit.

Neon supports four reasoning levels: low, medium, high, and xhigh. The high setting is selected by default to balance depth against latency and token consumption.

Google recommends evaluating all four levels to find the lowest setting that satisfies workload requirements. The xhigh setting is reserved for multi-step problems where high falls short.

Multimodal Input with Text-Only Output; Cyber Variant Coming

The model accepts text, images, video, audio, and PDFs, but generates only text output. It supports code execution, computer use, function calling, file search, structured outputs, URL context, and grounding through Google Search and Google Maps.

Google is also preparing a specialized variant called Gemini 4 Neon for Cyber. That version targets cybersecurity defense and will reach trusted defenders through the Fairwind Program, Google Discovery reported.

Migration Requirements for Developers

Migrating from Carbon, Argon, or Hydrogen requires no schema modifications. However, developers must adjust reasoning defaults, as Neon starts at high and supports xhigh, while Carbon and Argon default to medium.

Moving from Gemini 3 requires removing explicit cache fields and sampling parameters. Client-managed conversation histories must preserve prior interactions and reasoning signatures.

Common Questions About Gemini 4 Neon

What is the default reasoning level for Gemini 4 Neon?

The model defaults to the high reasoning level, which balances depth with latency and token consumption. Developers can manually adjust the setting to low, medium, or xhigh.

How does implicit caching work with Neon?

Implicit caching is enabled by default on prompts containing at least 4,096 tokens. It charges $0.15 per million stored input tokens, providing a 95% discount.

Which platforms support Gemini 4 Neon at launch?

The model is available through the Gemini API, Google AI Studio, Google Antigravity, and enterprise offerings. Subscribers to AI Pro and Ultra plans can access it in the Gemini application.