Google is rolling out Gemini 4 Neon, an AI model designed for coding, research and other complex tasks. The company says Neon combines advanced reasoning with faster and less expensive execution than larger models.
Subscribers to Google AI Pro and Ultra can select Neon in the Gemini app under “Deeper analysis.” It is also available through Google’s developer tools and Gemini for business. Google positions it between Hydrogen, its model for everyday tasks, and Carbon, which is intended for the most demanding projects.
In the Gemini app, Neon can work with uploads such as spreadsheets, PDFs and photos. Google says it can turn dense material into summaries and reports, use information from connected Google apps such as Gmail and Docs, and draw on web sources for current research. These are company claims, and users should still check important results and sources.
According to 9to5Google, Neon replaces Argon in the app’s model picker, while Gemini 3.1 Pro is being removed. Paid users will also see an “Auto” option alongside Hydrogen, Neon and Carbon.
Performance and pricing
Google points to a range of benchmark results to support its performance claims. Neon placed third on the Artificial Analysis Intelligence Index at its highest reasoning setting and second on the Vals Index, according to the company. It also ranked first on tests for legal research and tax work. Such results offer a comparison under test conditions, but they do not guarantee the same performance on individual assignments.
The model has a context window of one million tokens. That is the amount of material it can process in a single conversation or task, measured in units of text. It also offers four reasoning levels, allowing developers to choose how much effort the model spends on a request.
For API users, Neon costs $3 per million input tokens and $15 per million output tokens. Reused input from a cache costs $0.15 per million tokens. Google says Neon costs 32% less per task than Carbon on an Artificial Analysis comparison at Neon’s default high reasoning level. Actual costs will depend on the task and the amount of text processed.
Safety and availability
Google says Neon can help find, test and fix software vulnerabilities. A version for cybersecurity defenders will become available to eligible members of its Fairwind Program over the coming days.
The company also reports improvements in instruction following and lower rates of fabricated results in internal testing. In one external prompt injection benchmark, Google says the model resisted all tested attacks when additional safeguards were active at its highest reasoning level. Those results do not mean the model is immune to attacks in other situations.
Sources
- Introducing Gemini 4 Neon – Google
- Google launches Gemini 4 Neon with Carbon quality at Flash speeds, live for AI Pro – 9to5Google
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox: