Google Launches Gemini 4 Neon: Carbon-Level Reasoning at Flash Speeds

Admin


Google has rolled out Gemini 4 Neon, a new model that offers what the company calls frontier-level reasoning for complex, high-frequency work, but at the speed and cost of a much lighter model. Announced on November 18, Neon sits between Hydrogen and Carbon in the Gemini 4 family: Hydrogen handles everyday help, Carbon tackles the hardest long-horizon problems, and Neon is positioned as the balance of depth and efficiency. It is available now in the Gemini app for AI Pro and AI Ultra subscribers, and in Google Antigravity for paid subscribers.

Neon's headline claim is that it matches or beats Carbon on coding, reasoning and agentic tasks while running at Flash speeds. It carries the same 1 million-token context window as Hydrogen, with four thinking levels — Low, Medium, High and Extra High — and Google says it writes more clearly, reads uploads like spreadsheets, PDFs and photos more precisely, and pulls in relevant details from Google apps such as Gmail and Docs. Multi-step tasks in Gemini now run on Neon, which Google says completes complex jobs faster while stretching usage limits further.

The benchmark sheet is unusually strong for a mid-tier model. Neon ranks third overall with a score of 61 on the Artificial Analysis Intelligence Index at its highest thinking level, and first overall on the Artificial Analysis Coding Agent Index at 58.6. It also debuted second globally with a score of 1336 on the new Agent Arena Index across more than 4,200 real-world agent sessions, and took first place on both the Legal Research Bench and the Tax Agent Bench. It scored 73.5% on Chartography for reading dense professional charts and 75.8% on Agents' Last Exam v1.1. On pricing, Neon costs $3 per million input tokens and $15 per million output tokens — twice the price of Hydrogen. Argon and Carbon currently cost $2 and $10 until November 30, after which they rise to $4 and $20, leaving Neon 25% cheaper than Carbon. The launch also simplifies the Gemini model picker to Auto, Hydrogen for everyday help, Neon for deeper analysis, and Carbon for code and complex work — with Argon replaced by Neon and the older 3.1 Pro going away.

Why it matters

Neon is interesting not because it is the smartest model Google has ever shipped, but because it attacks the trade-off that has defined this whole generation of AI products: you could have fast and cheap, or smart and slow, but rarely both. A model that genuinely delivers Carbon-class reasoning at Flash speeds changes the economics of every product built on top of it. Agentic workflows — the multi-step jobs where models call tools, check their own work and keep going — are exactly where latency and cost per step decide whether a feature ships or stays a demo.

The pricing tells the same story. Google is not positioning Neon as a budget option; at twice Hydrogen's price it is a premium product. The bet is that customers will pay more per token because each task needs fewer tokens and less waiting. And the benchmark pattern is revealing: Neon's best results cluster in agent evaluations — Vending-Bench 2, AutomationBench, the Agent Arena Index — rather than in pure chat. Google is aiming squarely at the world where models do work, not just answer questions. If those scores hold up outside Google's own testing, Neon could become the default workhorse of the Gemini lineup, with Carbon reserved for the problems that truly need it.

Source: 9to5Google — Google launches Gemini 4 Neon with Carbon quality at Flash speeds, live for AI Pro (November 18, 2026). Summary rewritten by VCTRIA.

Post a Comment