Back to Blog
    AI & AutomationNov 19, 2026

    Gemini 4 Neon: Google's Fast, Cheap Frontier Model for Business

    Amir

    Amir Farid

    7 min read

    Gemini 4 Neon: Google's Fast, Cheap Frontier Model for Business

    Google launched Gemini 4 Neon on 18 November 2026. It sits between Gemini 4 Hydrogen, the fast everyday model, and Gemini 4 Carbon, the frontier heavyweight — built for complex software engineering, knowledge work and science. The API costs $3 per 1M input tokens, $0.15 per 1M cached input, and $15 per 1M output tokens, with a 1M-token context window. Google says Neon matches or beats Carbon in quality while being 60% cheaper and 3x faster — Google's claim, not an independent result.

    The short version: for coding agents, long documents and multi-step knowledge tasks, Neon is now the default to pilot. Carbon stays for the hardest problems, Hydrogen for the cheapest volume. Below: the prices, the benchmark claims, and how to route work between the models. All launch facts are from Google's own announcement.

    Gemini 4 Neon: Google's Fast, Cheap Frontier Model for Business

    What Gemini 4 Neon actually is

    Neon is the third member of the Gemini 4 family, for complex software engineering, knowledge work and science — work where a model plans, uses tools, checks its own output and keeps going across many steps. It is adapted from Gemini 4 Carbon, with a March 2026 knowledge cutoff, and four configurable thinking levels let you trade reasoning effort against speed and cost.

    Where it sits in the family

    Think of three rungs. Hydrogen is the fast, cheap rung for everyday volume; Carbon is the top rung for the hardest problems; Neon is the middle rung that does serious work at a lower price. Our Gemini 4 Hydrogen guide covers the cheap end, and our Gemini 4 Carbon guide covers the top end.

    It is rolling out from 18 November in the Gemini API in Google AI Studio, Gemini for business, Google Antigravity and the Gemini app for paid Google AI Pro and Ultra subscribers — and generally available in GitHub Copilot. A cyber-focused variant is also rolling out to vetted defenders through Google's Fairwind Program.

    The price: what $3 and $15 actually mean

    Laptop with an abstract code editor and a glowing Gemini spark on a desk

    Input costs $3 per 1M tokens, cached input $0.15 per 1M — a 95% discount — and output $15 per 1M. The cache discount is the number to watch: agent work re-reads the same documents and codebases on every step, and cached input at $0.15 is where long-running jobs get cheap.

    The cost maths, in plain numbers

    • A 1M-token input with half cached costs about $1.65.
    • Add 200k output tokens and the call costs about $4.65 in total.
    • On a long agent run, the effective input price trends toward $0.15 per 1M.

    Compare that with Carbon's pricing in our Carbon guide. Note the Argon pricing deadline: Argon's introductory $2/$10 pricing ends on 30 November 2026, rising to $4/$20. That deadline applies to Argon only — Google states no expiry date for Neon. Price the whole workflow, not the single call, the same discipline behind any serious AI automation project.

    What the benchmarks claim — and what they don't prove

    Every score here is Google's own claim, not an independent result. At its highest thinking level, Google says Neon scores 61 and ranks #3 overall on the Artificial Analysis Intelligence Index, #2 on the Vals Index, and #1 on Artificial Analysis's Coding Agent Index at 58.6 — Google claims the best coding-agent score of any model. A #1 there is a reason to pilot Neon on your own repositories, not a guarantee it beats Carbon on your hardest problem. Our GDPx agent benchmark post showed even the best AI agent scores only around 12% on long-horizon business tasks. Measure success rate with a human reviewer, cost per completed task, and how often staff must intervene.

    Neon vs Carbon vs Hydrogen: which model for which job

    The three models are a routing problem, not a ranking. Carbon is the strongest and most expensive; Hydrogen is the fastest and cheapest; Neon is the middle that Google says matches or beats Carbon in quality at 60% lower cost and 3x the speed. If that claim holds on your tasks, Neon replaces Carbon for most serious work — but "if" is doing the work, so test it.

    A simple routing rule

    • Hydrogen first: drafting, summaries, triage — anything a human reviews quickly.
    • Neon for the serious middle: coding agents, multi-step analysis, long documents, science and knowledge work.
    • Carbon when it matters most: the hardest problems and final review of high-stakes work.

    Escalate on failure: if Neon fails a check twice, re-run the step on Carbon. The same routing logic applies across vendors — see our GPT-6.2 Astra and Sol guide, our Claude Fable 5.2 guide, and our Qwen4 Max guide.

    Where Neon earns its money first

    Neon's best first use is coding agents. It is generally available in GitHub Copilot and built into Google Antigravity, so developers can try it inside tools they already use, on real repositories, with a human reviewing every change. Start with contained work: bug fixes, test generation, refactoring a single module.

    Knowledge work is the second win

    The 1M-token context window makes Neon practical for knowledge work: whole contracts, long reports, research spanning many sources. With the 95% cache discount, re-reading a 200-page document on every agent step no longer dominates the bill. That is the same pattern behind good AI agents for small business automation. Do not put Neon on trivial volume work — Hydrogen is cheaper there.

    How to pilot Neon in two weeks

    Pick one routine workflow and one hard workflow. Run the routine one on Neon for two weeks, and test the hard one on Neon against Carbon side by side. Keep a human in the loop on every client-facing output, and log pass rate, cost and fix-up time per task.

    A practical routing plan

    • Week 1: coding agent pilot in Copilot or Antigravity on one repository.
    • Week 2: compare Neon against Carbon on hard tasks, and Hydrogen on routine ones.
    • Then decide: Neon for coding and knowledge work, Hydrogen for volume, Carbon for the top tier.

    Four thinking levels give you a further dial: a low level for fast first passes, the highest level for work that must be right. If you need help building the agent around the model, our AI agents development team builds on exactly this kind of routing — see our AI agent development cost guide for UK pricing.

    The catches: no free tier, a March cutoff, and cost discipline

    Three things to know before you commit. First, there is no free tier on the API, as reported — every call is billed, so set usage limits and alerts before you hand Neon to a team. Second, the knowledge cutoff is March 2026, so anything newer needs retrieval. Third, the 60% cheaper and 3x faster figures are Google's claims against Carbon; your own pilot is the only number that counts. Our custom software development cost guide shows how model costs sit inside a full build budget, and if the pilot proves the concept, custom software development or a Flutter app can carry it to production. The same routing applies to voice agents and ecommerce agents — and if you want to talk through your pilot, contact Coheser.

    Frequently asked questions

    When did Gemini 4 Neon launch, and where is it available?

    18 November 2026, in the Gemini API in Google AI Studio, Gemini for business, Google Antigravity and the Gemini app for paid Pro and Ultra subscribers, plus GitHub Copilot. No free tier on the API, as reported.

    How much does Gemini 4 Neon cost on the API?

    $3 per 1M input tokens, $0.15 per 1M cached input (a 95% discount) and $15 per 1M output tokens. Google says it matches or beats Carbon in quality while being 60% cheaper and 3x faster — Google's claim. No pricing expiry has been announced; the 30 November deadline applies to Argon only.

    What is Gemini 4 Neon best used for?

    Complex software engineering, knowledge work and science — especially coding agents and long-document analysis, with a 1M-token context window and four thinking levels. Routine volume work is still cheaper on Hydrogen; the hardest problems still belong on Carbon.

    Should my business switch to Gemini 4 Neon now?

    Pilot first. Google claims Neon ranks #1 on the Coding Agent Index and #3 overall on the Artificial Analysis Intelligence Index, but those are Google-reported scores, not independent results. Run a two-week pilot and scale only what the evidence supports.

    #Gemini 4 Neon#Google Gemini#AI Agents#AI Automation