Photo: Google
MOUNTAIN VIEW, November 19, 2026 — Google has a new middle child, and it is aiming at the top of the class. Google DeepMind has introduced Gemini 4 Neon, a model built for complex coding, analysis and science work that it pitches as frontier-level intelligence at a lower cost per task than its flagship.
Announced on Wednesday (18 November) in a blog post by Koray Kavukcuoglu, Google DeepMind’s senior vice-president and Google’s chief AI architect, Neon sits between the two existing members of the Gemini 4 family. Carbon handles the hardest long-horizon projects, Hydrogen covers fast everyday execution, and Neon is meant to balance depth with efficiency for complex, high-frequency workflows.
Price and context window
Neon is priced at US$3 per million input tokens (US$0.15 per million cached input tokens, a 95% discount) and US$15 per million output tokens, with a one-million-token context window. Google says that at its default “high” thinking level, Neon costs 32% less per task than Carbon. The model has four thinking levels, letting teams trade intelligence against token spend.
What Google says it can do
According to Google, Neon ranks third overall with a score of 61 on the Artificial Analysis Intelligence Index at its “xhigh” thinking level, and first on the Artificial Analysis Coding Agent Index with 58.6. It reports 65.2% on Terminal-Bench 4.1, second place on the Vals Index (70.9%) and first on Zapier’s AutomationBench (56.3%).
The company is pitching Neon at long-running agent workflows, either on its own or as an orchestrator directing parallel Hydrogen sub-agents in Google Antigravity, its agentic coding environment. Google says the model cuts redundant file reads, handles context compaction reliably in long sessions and recorded its lowest fabrication rate yet in internal Antigravity testing.
These are Google’s own benchmark claims, and independent testing will show how they hold up in day-to-day use.
A version for cyber defenders
Google is also releasing Gemini 4 Neon for cyber, aimed at finding, validating and patching code vulnerabilities. It says the model reached 90.5% on SEC-bench Pro and 51.26% on WebExploitBench v1.2. That version is rolling out to trusted defenders in the coming days through Google’s Fairwind Program.
On safety, Google says Neon keeps the same protections against severe misuse and prompt injection as Carbon, paired with real-time safeguards against cyber and biosecurity misuse.
Where to get it
Neon is rolling out in the Gemini API through Google AI Studio, in Gemini for business, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers, who can pick it from the model menu. Google says Tasks in Gemini can now also run on Neon.
Source: Google.

