Gemini 4 Neon launches: Carbon quality at Flash speeds, for AI Pro

Following the launch of Carbon and Hydrogen last week, Google today launches Gemini 4 Neon. The model offers “frontier-level reasoning for faster, more cost-effective, complex, high-frequency workflows.” Before today, Google’s Gemini 4 family consisted of: Hydrogen: Everyday Help: Flash replacement Argon: Code and Analysis: “…ideal choice for most knowledge and software engineering work” Carbon: More complex work: “…built for tasks that test the limits of a model’s intelligence” Neon performs comparable, if not superior, to Carbon in coding, reasoning, and agency tasks, but at Flash speeds and at a lower cost. As such, Neon is available in the Gemini app for AI Pro (and AI Ultra) subscribers, and 3.1 Pro will eventually go away. Neon also replaces argon. Meanwhile, 3.5 Flash-Lite is now “Auto” in the model selector for paid users. Advertisement: Scroll for more content With this release, the model selector is: Automatic: Adapts to your needs Daily Help: 4 Hydrogen Deeper Analysis: 4 Neon Code and complex work: 4 Carbon In the Gemini app, Google says to expect: Higher-quality writing and knowledge work: Neon writes clearly, with well-structured responses and a natural, direct tone. Reads uploads like spreadsheets, PDFs, and photos with greater accuracy, turns dense material into useful reports and summaries, and extracts relevant details from Google apps like Gmail and Docs. Trusted research and analysis: Neon frequently pulls in details from sources around the web to keep answers up-to-date on fast-moving topics, like comparing prices between vendors before a big purchase or putting together a complex summary of the latest market data. Faster help for more difficult tasks: Tasks on Gemini can now be run on Neon, performing complex, multi-step jobs more quickly and efficiently than with larger models. Get more done while pushing your usage limits even further by completing tedious to-do tasks, like analyzing account statements to pinpoint why a utility charge skyrocketed or checking multiple receipts and documents to negotiate a refund. Neon has the same 1 million token context window with four levels of thought, like Hydrogen: Low, Medium, High and Extra High. Officially, the new model sits between Hydrogen and Carbon in the Gemini 4 family. Neon is designed for complex coding, analytical reasoning, and production-level enterprise workflows. While Carbon tackles the most difficult, long-term projects and Hydrogen takes care of fast daily execution, Neon balances depth and efficiency. In terms of benchmarks: Neon ranks ahead of comparable models and ranks #3 overall with a score of 61 on the Artificial Analytics Intelligence Index at its x-high thinking level, ranking just behind larger, slower models. Neon is ranked #2 in addressability (clean adaptation to user corrections mid-task) in the new Agent Arena Index. …reaches the number 1 position in both the legal research bank and the tax agents bank. Neon also shows strong wins across all categories in Surge AI, ranking #1 in most benchmarks in finance, accounting, spreadsheet modeling, and complex constraint tracking. It also scores 73.5% in Chartography for reading dense professional charts… …performs on par with leading models in the latest Agent Exam v1.1 (75.8%). It ranks #1 overall in both Zapier’s AutomationBench (56.3%) and Andon Labs’ Vending-Bench 2 ($16,843), a long-term simulation of running a business. …debuts at #2 globally with a score of 1,336 on Arena’s new Agent Arena Index across over 4,200 real-world agent sessions. Neon is best in class in Terminal-Bench Science (70.5%), RiemannBench (84.0%), and HLE-Diamond (71.3%). …ranking #1 overall in the Artificial Analysis Coding Agent Index (58.6) with a score of 65.2% in Terminal-Bench 4.1 and 46.2% (at its default high setting) in FrontierCode 2.0. Neon is priced at $3/1 million input tokens and $15/1 million output tokens, or twice as expensive as hydrogen. Argon and carbon will cost $2 per 1 million input and $10 per 1 million output until November 30. After the promotional price, it will be $4 for 1 million input and $20 for 1 million production, making neon 25% cheaper as a result. The model is also available on Google Antigravity for paid AI Pro or Ultra subscribers. FTC: We use automatic affiliate links that generate income. Further.