Google is expanding the Gemini 4 family with Neon. The model is meant to combine the intelligence of a top-tier model with fast execution of tasks in software development, knowledge work and science.
Google is introducing Gemini 4 Neon. It is already the fourth model Google has launched within the Gemini 4 family in a single month. Google is working its way further through the periodic table and has now landed on neon. The Neon model sits between Gemini 4 Carbon, for the heaviest long-running projects, and Gemini 4 Hydrogen, for fast everyday work.
The model has a context window of one million tokens and four thinking levels. Developers use these to decide per task how much reasoning and how many tokens they spend. At the default level, high, the cost per task is 32 percent lower than with Carbon, according to Google.
Gemini 4 Neon benchmarks
At the highest thinking level, Neon scores 61 points on the Artificial Analysis Intelligence Index. That is good for third place. On the Artificial Analysis Coding Agent Index, Neon ranks first with 58.6 points. On Terminal-Bench 4.1, the model scores 65.2 percent.
With Neon, Google is targeting AI agents that work for long periods and with many tools. In the Google Antigravity development environment, Neon can act as an orchestrator that directs parallel subagents running on Hydrogen. According to Google, Neon made up fewer facts in internal tests than any other model the company has built so far.
A version for cyber defence is also coming. According to Google, Neon writes more secure code than Carbon. Members of the Fairwind Program, through which Google gives selected cyber defenders early access, will get access in the coming days.
From Argon to Neon
Google released the four Gemini 4 models in seven weeks. Argon came first on 30 September, aimed at complex, long-running tasks and with output of up to one million tokens. On 28 October came Carbon, which Google calls its strongest model, with an emphasis on cybersecurity and long-running tasks. Carbon initially went only to cyber defenders and has been widely available since 10 November.
Hydrogen appeared on 12 November as a fast, cheap model for everyday programming work and AI agents. It runs in Google Search and is the only Gemini 4 model that can also be used for free via the Gemini API. With the launch of Neon, Google now describes the division of tasks only in terms of Carbon, Neon and Hydrogen. Argon remains available via the Gemini API.
New prices on 1 December
Neon is available from now on via the paid Gemini API in Google AI Studio and Android Studio, in Gemini for business and in Antigravity for Google AI Pro and Ultra subscribers. Via the Gemini API, the price is 3 dollars per million input tokens and 15 dollars per million output tokens. For cached input, you pay 0.15 dollars.
Argon and Carbon cost 2 and 10 dollars during their introductory period, which ends on 30 November. Until then, Neon is the most expensive Gemini 4 model per token. From 1 December, Argon and Carbon rise to 4 and 20 dollars, and Neon is a quarter cheaper per token than both. At 1.50 and 7.50 dollars, Hydrogen costs exactly half as much as Neon.
