Google released Gemini 4 Neon today, Nov. 18, a new artificial intelligence model built for coding, knowledge work and long-running AI agents. It is rolling out in the Gemini API in 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. Developers pay $3 per million input tokens and $15 per million output tokens.
Neon is the fourth model in Google’s Gemini 4 family, following Argon, Carbon and Hydrogen, and it arrived barely a month after the lineup began. Google is pitching it as a model people will use every day, landing between the cheaper Hydrogen and the flagship Carbon.
What can Gemini 4 Neon do?
Neon takes in text, images, video, audio and PDFs, but it only produces text. It has a context window of about 1 million tokens and supports computer use, code execution, file search and grounding through Google Search and Maps. It offers four thinking levels, so users can trade speed for deeper reasoning.
Google says Neon was built for long-running agent work involving many tools, either on its own or as an orchestrator directing parallel Hydrogen subagents in Google Antigravity. A separate version, Gemini 4 Neon for cyber, is rolling out to trusted security defenders in the coming days through Google’s Fairwind Program.
Starting today, Gemini 4 Neon is available to Google AI Pro and Ultra subscribers on mobile, desktop, and web.
This new model excels at work that requires frontier-level reasoning and fast responses. 🧵
— Google Gemini (@GeminiApp) November 18, 2026
How does Gemini 4 Neon perform on benchmarks?
Google’s own numbers are strong. It says Neon ranks first on the Artificial Analysis Coding Agent Index at 58.6, first on Zapier’s AutomationBench at 56.3% and second on the Vals Index at 70.9%. Google also says Neon ranks first on Legal Research Bench and Tax Agent Bench. The company reports a score of 61 on the Artificial Analysis Intelligence Index at its highest thinking level, which would place it third overall.
Context matters here. According to Artificial Analysis, Anthropic’s Claude Fable 5.2 scores 65 on that index and OpenAI’s top Astra model scores 62, both above Neon, while Gemini 4 Carbon sits at 60. Google’s launch table shows Neon ahead of Claude Opus 5.5 and OpenAI’s Sol model on most of the tests it lists, but those numbers come from Google, and independent evaluations have not yet arrived.
Is Gemini 4 Neon actually cheaper?
It depends on how you count. Google calls Neon ‘Flash tier’ in cost, but at $3 and $15 per million tokens, it costs twice as much as Hydrogen, which runs $1.50 and $7.50. It also costs more than Carbon’s introductory price of $2 and $10, though Carbon’s rate doubles to $4 and $20 on Dec. 1.
Google’s argument is about the total bill, not the per-token rate. The company says Neon’s cost per task at the high thinking level is 32% lower than Carbon’s. Models that think longer burn more tokens, so the real cost depends on how much reasoning each job needs. For startups and independent developers watching every dollar, that difference between price per token and price per task is the number to test.
Why did Gemini 4 Neon show up before Google announced it?
Watchers spotted it early. A reference to the model appeared in the Gemini web app on Nov. 16, then in Google Cloud’s pricing catalog and in Antigravity’s command line tool on Nov. 18, before Google made it official the same day. Some had guessed Neon would be a small, budget model. It turned out to be a midweight reasoning model.
The pace is the bigger story. Google has stacked Gemini 4 releases week after week, and Neon will not be the last. Rumored names like Helix and Krypton are already circulating. The race among Google, OpenAI and Anthropic is moving faster than most businesses can test the models.

