Add as a preferred source on Google
X
Mail

Between Carbon and Hydrogen: “Gemini 4 Neon” Now Available

Gemini 4 Neon
この記事のポイント
  • Google has announced “Gemini 4 Neon,” a new AI model in the next-generation Google AI model family “Gemini 4,” now rolling out to developers, enterprises, and “Google AI Pro” and “Google AI Ultra” members.
  • It is positioned between “Gemini 4 Carbon,” the top-tier model built for the most difficult and long-running projects, and “Gemini 4 Hydrogen,” which handles everyday tasks at high speed.
  • Supports four “thinking level” options: “Low,” “Standard,” “High” (default), and “Ultra/Maximum.”

On Wednesday, November 18, 2026 (local time), Google announced “Gemini 4 Neon,” a new AI model from the next-generation Google AI model family “Gemini 4,” and made it available to developers, enterprises, and members of “Google AI Pro” and “Google AI Ultra.”

“Gemini 4 Neon” is an AI model positioned between “Gemini 4 Carbon“—the top-tier model designed for the most difficult and long-term projects in the “Gemini 4” series—and “Gemini 4 Hydrogen,” which handles everyday tasks at high speed. Combining frontier-level intelligence with fast, cost-efficient, end-to-end task completion capabilities, it is designed for complex software engineering, knowledge work, and scientific workflows.

It has also recorded high performance across various benchmarks. On the “Artificial Analysis Intelligence Index,” it ranked 3rd overall with a score of 61 at the highest “xhigh” thinking level setting, and secured 2nd overall (70.9%) on the “Vals Index.” Furthermore, it achieved 1st place overall on the “Legal Research Bench,” “Tax Agent Bench,” Zapier’s “AutomationBench” (56.3%), and “Vending-Bench 2” ($16,843).

In coding and agent workflows, it functions as the “Orchestrator” for “Google Antigravity,” managing multiple sub-agents (Gemini 4 Hydrogen). On its own, it also achieved top-tier scores, including 1st place overall on the “Artificial Analysis Coding Agent Index” (58.6), 65.2% on “Terminal-Bench 4.1,” and 39.0% on “SWE-sweep” (in the Google Antigravity environment). In addition, its 1-million-token context window allows for the integrated analysis of diverse inputs such as spreadsheets, PDFs, charts, audio, and video within a single session.

Improvements have also been made in safety and reliability, recording the lowest hallucination rate to date in internal “Google Antigravity” testing. On the “Gray Swan IPI” benchmark measuring resistance to indirect prompt injection attacks, it achieved a 0.0% attack success rate through the combination of its highest thinking level and real-time protection features.

Like other AI models, “Gemini 4 Neon” supports four “thinking level” options: “Low,” “Standard,” “High” (default), and “Ultra/Maximum.” Its default “High” setting reduces task-based costs by 32% compared to “Gemini 4 Carbon” while maintaining high intelligence, and even its highest setting achieves a cost per task that is over 36% lower than competing frontier models.

“Gemini 4 Neon” is now available through “Google AI Studio,” “Android Studio,” and the paid tier of the “Gemini API” via “Gemini for business” with a 1-million-token context window, priced at $3 per 1 million input tokens (cached input at a 95% discount of $0.15) and $15 per 1 million output tokens. It is also available for “Google AI Pro” and “Google AI Ultra” members via “Google AI Studio,” “Google Antigravity,” and the model selection dropdown in the “Gemini” app.

Additionally, “Gemini 4 Neon for cyber,” specialized for cyber defense, is scheduled to launch within days through the “Fairwind Program,” a trusted defense agency program.

Source:Google(https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/)


*This site uses affiliate advertising.