Back home

Gemini 4 Neon

sitemap+x+web 0.95
Visit source

Intro

Gemini 4 Neon is the Gemini 4 family model Google officially introduced on 18 November 2026. It sits between already-filed Gemini 4 Carbon and Gemini 4 Hydrogen. In the official post Introducing Gemini 4 Neon, Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, wrote: “Today, we’re introducing Gemini 4 Neon — bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.”

The Gemini 4 Neon model card (published 18 November 2026) adds: “While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency.” It also states that “Gemini 4 Neon is adapted from Gemini 4 Carbon.” The knowledge cutoff is March 2026.

At 21:34:04 UTC the same day, @GoogleDeepMind posted: “Gemini 4 Neon is rolling out now. Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science.” At 21:37:20 UTC, @GeminiApp said it was available to Google AI Pro and Ultra subscribers in the Gemini app. This slot’s sitemap crawl added the official blog and model card; Cursor docs and OpenRouter pages are availability follow-ons, not the first introduction.

This entry records the official model name Gemini 4 Neon. Do not merge it with Gemini 4 Carbon, Gemini 4 Hydrogen, Gemini 4 Argon, or DeepPrinciple’s Hydrogen-1.0. Earlier radar slots had seen unofficial leaks or directory rumors and did not file the name; the dated official blog on 18 November 2026 is the first formal introduction.

Advertisement

Explain

Officially, Neon sits in the middle of the Gemini 4 family: Carbon is for the hardest, longest-horizon projects, Hydrogen is for fast everyday execution, and Neon is meant to balance depth and efficiency on complex, high-frequency workflows. Pricing on the blog is $3 per million input tokens, $0.15 per million cached input tokens (a 95% discount, Google’s figure), and $15 per million output tokens, with a 1 million-token context window. At its default high thinking level, Google says Neon pairs a high Artificial Analysis Intelligence Index score with 32% lower cost per task than Carbon. The xhigh thinking level is where it posts its highest intelligence score, still described as 36%+ cheaper per task than competing models at their highest setting. These are company claims, not independent reruns by this radar.

Figures stated in the official blog include: a score of 61 and #3 overall on the Artificial Analysis Intelligence Index at xhigh; #2 on the Vals Index (70.9%), described as being on the Pareto frontier; #2 on steerability and a global score of 1336 on the new Agent Arena Index (4,200+ real-world agent sessions). For coding, Google says Neon ranks #1 overall on the Artificial Analysis Coding Agent Index (58.6), scores 65.2% on Terminal-Bench 4.1 and 46.2% at default high on FrontierCode 2.0, and on SWE-sweep leads the mini-SWE harness at 12.8% and reaches 39.0% in Google Antigravity. For science and math, the blog lists Terminal-Bench Science 70.5%, RiemannBench 84.0%, and HLE-Diamond 71.3%. The model card also tabulates some of these results next to Claude Opus 5.5 and GPT-6.2 Sol, and lists CursorBench 4.1 at 65.4% (58.1% for Claude Opus 5.5 in that table). Cite the 18 November 2026 blog or model card when repeating any of these numbers.

In Google Antigravity, the blog says Neon can run long, tool-heavy work on its own or act as a fast, high-judgment Orchestrator directing parallel Hydrogen subagents. It also introduces “Gemini 4 Neon for cyber,” aimed at secure code generation, penetration testing, and vulnerability finding, validation, and patching, rolling out in the coming days to trusted defenders in the Fairwind Program. On safety, Google says Neon builds on the Gemini 4 safety and security stack and keeps Carbon-level protections against severe misuse and prompt injection; the full Safety & Security Report is described as coming in the following days.

Availability follows the blog’s “Start building” section: developers can use the Gemini API paid tier in Google AI Studio and Android Studio, and paid Pro or Ultra subscribers can try Neon in Antigravity and AI Studio; enterprises can use it in Gemini for business (API or model dropdown) and Antigravity; consumers who are Google AI Pro or Ultra subscribers can select it in the Gemini app; cyber access is requested through the Cloud account team for the Fairwind Program. Inputs include text, images, audio, and video. The model card describes a combined input-output context of up to 1M tokens, with up to 64K default synchronous output tokens and up to 1M via background long decode.

Usage

  1. Read the official blog at https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/ , and cross-check the model card at https://deepmind.google/models/model-cards/gemini-4-neon/ , the @GoogleDeepMind announcement at https://x.com/GoogleDeepMind/status/2123167884411077030 , and the @GeminiApp post at https://x.com/GeminiApp/status/2123168704662012079 ;
  2. Per Google’s rollout note, select Gemini 4 Neon in the Gemini API paid tier in Google AI Studio, Gemini for business, Google Antigravity for paid Google AI Pro or Ultra subscribers, or the Gemini app model picker for those subscribers. Cursor’s model page is https://cursor.com/docs/models/gemini-4-neon and OpenRouter lists https://openrouter.ai/google/gemini-4-neon ; availability follows each product’s current page;
  3. When citing Intelligence Index, Vals, Terminal-Bench, pricing, cost-per-task, or CursorBench figures, attribute them to the 18 November 2026 official blog or model card and re-check the source;
  4. Do not treat Gemini 4 Neon as Gemini 4 Carbon, Gemini 4 Hydrogen, Gemini 4 Argon, or DeepPrinciple’s Hydrogen-1.0, and do not treat “Neon for cyber” or the Fairwind Program as a separate foundation model. This page only restates the official blog, model card, and announcement posts.