Google has launched a new AI model called Gemini 4 Neon, and it is aimed squarely at the people who use AI for real work rather than quick questions. Announced on Wednesday, November 18, 2026, Neon is designed to finish complex, multi-step jobs from start to end — writing and fixing code, analysing dense documents, building financial models, and even hunting security flaws — while running about three times faster than Google’s top model and costing roughly 60 per cent less.
Neon is the third member of Google’s Gemini 4 family to reach the public in this cycle, and it sits in a deliberately chosen middle ground. It is not the biggest model Google has, and it is not the fastest lightweight one either. It is the one Google wants developers, businesses, researchers and security teams to reach for every day, for the heavy jobs they repeat again and again.
Here is everything that matters about Gemini 4 Neon: what it does, how fast and how cheap it really is, how it scores against rival models, and — importantly — who can actually use it, because free Gemini users are left out.
What is Gemini 4 Neon?
Google describes Gemini 4 Neon as a model built for “fast end-to-end task completion” across complex software engineering, knowledge work and science workflows. That phrase matters, because it is a different job from answering a single question. Neon is built to take a messy pile of inputs — spreadsheets, PDFs, charts, audio and video — work through them in several steps, use tools along the way, and hand back a finished deliverable such as a report, a summary or an analytical output.
It can process up to 1 million tokens of information at a time. In practical terms, that means it can hold very large documents, long codebases or hours of material in its working memory at once, instead of forcing you to chop the work into small pieces.
Inside the Gemini app, Google says Neon reads uploaded files more precisely than earlier models, prepares reports and summaries from them, and can draw relevant information from your Google apps, including Gmail and Google Docs. The idea is that a researcher can drop in a stack of papers, a developer can point it at a repository, and a business user can feed it financial records — and get structured work back rather than a chat reply.
Google also says Neon recorded its lowest fabrication rate so far in internal testing inside Google Antigravity, its agentic development environment. In other words, it is supposed to make up fewer facts than earlier models when it works through long, multi-step jobs — a claim that matters a lot when the output is a tax report or a security patch rather than a poem.
Where Neon fits in the Gemini 4 family
Google names its Gemini 4 models after chemical elements, and each one has a clear role. Understanding the family is the fastest way to understand Neon, because Google is not selling it as “the best model” — it is selling it as the right model for a specific kind of work.
Gemini 4 Carbon is the heavyweight. It is built for the hardest, longest-horizon projects, where maximum reasoning depth matters more than speed or cost. Gemini 4 Hydrogen, launched earlier in November, is the sprinter — designed for fast, everyday execution and quick answers. Neon sits between the two: it balances depth and efficiency for complex, high-frequency workflows.
That middle position is the whole point. Carbon is overkill — and too slow and expensive — for a job you run fifty times a day. Hydrogen is fast, but it is not built for a multi-file debugging session or a 200-page document review. Neon is aimed at users who need serious reasoning and fast responses in the same model: developers managing large codebases, businesses processing dense documents and data, researchers handling technical analysis, and cybersecurity teams looking for faster ways to find and fix vulnerabilities.
The launch also reshuffles the Gemini app’s model picker. Alongside Neon, Google has replaced Gemini 4 Argon in the Gemini app and shuttered the Gemini 3.1 Pro model. If you open the model dropdown and Argon is gone, that is why — Neon is its replacement.

Gemini 4 Neon benchmarks: how it scores
Benchmarks never tell the whole story, and every one of these figures comes from Google’s own testing or from independent indexes Google chose to highlight — so treat them as a signal, not a verdict. Even so, the pattern is consistent: Neon scores near the top on the tests that measure real work rather than trivia.
On the Artificial Analysis Intelligence Index, an independent benchmarking platform, Neon ranks third overall with a score of 61 at its highest “xhigh” thinking level. It ranks second on the Vals Index and second for steerability on the new Agent Arena Index, which measures how well a model adapts when a user corrects it mid-task — a crucial skill for an agent that is supposed to work with you, not just for you.
For coding, the numbers are stronger still. Google says Neon ranks first overall on the Artificial Analysis Coding Agent Index with a score of 58.6. It also reports a 65.2 per cent score on Terminal-Bench 4.1 and a 46.2 per cent score on FrontierCode 2.0 at its default high setting.
For everyday and business automation, Neon claims the top spot on Zapier’s AutomationBench, which tests how well AI can automate routine work, and first place on Andon Labs’ Vending-Bench 2, a long-horizon simulation in which the model runs a virtual vending machine business. It also topped benchmarks for legal research and tax-related tasks, according to Google.
One early caution: independent coverage noted that Neon ranked fourth in a separate benchmark still under development, behind Fable 5.2, Astra and Carbon — but those results came after only hours of availability. Early leaderboard positions move fast, and the coding and automation results are the ones with more substance behind them so far.
| Benchmark | Gemini 4 Neon result |
|---|---|
| Artificial Analysis Intelligence Index | 3rd overall, score 61 (xhigh thinking) |
| Artificial Analysis Coding Agent Index | 1st overall, score 58.6 |
| Terminal-Bench 4.1 | 65.2% |
| FrontierCode 2.0 | 46.2% (default high setting) |
| Vals Index | 2nd |
| Agent Arena Index (steerability) | 2nd |
| Zapier AutomationBench | 1st |
| Andon Labs Vending-Bench 2 | 1st |
Gemini 4 Neon price: 3x faster, 60% cheaper than Carbon
The headline numbers Google wants you to remember are speed and cost. According to Google, Neon operates about three times faster than Gemini 4 Carbon while reducing overall costs by around 60 per cent. That combination — frontier-level capability at a much lower price — is what makes it a daily-driver model rather than a special-occasion one.
The actual pricing is straightforward. Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens through the Gemini API. Cached input tokens — previously processed information the model can reuse instead of reading again — cost just $0.15 per million, a 95 per cent discount. For developers running the same context through the model repeatedly, that caching discount is where the real savings add up.
For comparison, Carbon sits at the top of the price range because it is built for the hardest jobs. Neon undercuts it sharply while keeping, in Google’s words, frontier-level intelligence. Whether that trade-off holds up in independent testing is the question developers will answer over the next few weeks, but on paper the value proposition is clear: most of the capability, a third of the time, at roughly 40 per cent of the cost.
Built for coding and long agent workflows
Neon’s strongest pitch is to developers. Google says the model is engineered for long-running agent workflows — complex, tool-heavy tasks that run for a long time and involve many steps. It can work independently, or act as an orchestrator that directs parallel Hydrogen subagents inside Google Antigravity, splitting a big job into smaller pieces and coordinating the results.
For developers, that means multi-file debugging and refactoring across whole repositories through targeted edits. Google says Neon reduces repeated file reads, shares progress while it works, and manages context compaction reliably in long sessions — the unglamorous plumbing that decides whether an AI coding assistant stays useful after the first hour or slowly falls apart as the conversation grows.
It is also trained to complete multi-step work without cutting corners, fabricating results, or exceeding the boundaries a user sets. Google reports a 0.09 per cent misaligned action rate in its agentic safety evaluation and a 99.2 per cent pass rate on its agent honesty evaluation. Those are internal figures, but they point at the same goal: an agent you can leave running on a real task without it quietly doing something you did not ask for.
A special version for cybersecurity
A major part of the launch is Gemini 4 Neon for cyber, a version built to help security teams write, test and repair secure code. Google says it will be rolled out to trusted defenders through its Fairwind Programme in the coming days, and eligible defenders can request access by contacting their Google Cloud account team.
The cyber numbers are striking. Google says Neon produces more secure code than Gemini 4 Carbon across easy, medium and hard challenges on AutoBaxBench, reaching a state-of-the-art score of 76 per cent on the hard tier. It is also designed to identify vulnerabilities, validate which risks are actually exploitable, and patch security flaws. Google reports a 90.5 per cent score on SEC-bench Pro for vulnerability finding and a 70.7 per cent score on Collinear’s CWE-bench v2 for vulnerability patching.
That is a meaningful shift if it holds up. Most AI models can write code quickly; far fewer can write code that is actually secure, or reliably find and fix the holes in someone else’s code. A model that does both — and does it fast and cheaply — is exactly what security teams have been asking for, and it explains why Google is gating the cyber version behind a trusted-defender programme rather than releasing it to everyone at once.
Safety: strong guards, fewer refusals
Google is pitching Neon as a safer agent as well as a faster one. The model includes safeguards against severe misuse and prompt injection, while aiming to reduce unnecessary refusals of legitimate work — the frustrating “I can’t help with that” responses that safe models sometimes give to perfectly normal requests.
Google says Neon maintains Carbon-level protections against cyber and biosecurity misuse. On Gray Swan’s indirect prompt injection benchmark, Google says Neon resisted every attack at its highest thinking level when real-time safeguards were enabled. Indirect prompt injection — where malicious instructions are hidden inside a document or webpage the model reads — is one of the biggest real-world risks for AI agents that read files and browse the web, so a clean result there matters.
The balance Google is trying to strike is clear: a model that is hard to misuse, but not so cautious that it refuses real work. Whether that balance holds in practice will depend on how it behaves outside Google’s own evaluations, but the stated goal is the right one for a model designed to be left running on multi-step tasks.
How to access Gemini 4 Neon — and who can’t
Here is the part that will disappoint a lot of people: Gemini 4 Neon is not available on the free Gemini plan.
Developers can build with it through the paid tier of the Gemini API in Google AI Studio and Android Studio. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity, and Pro and Ultra subscribers can also select Neon from the model dropdown in the Gemini app. Enterprises can use it through Gemini for Business, either via the API or the model dropdown, as well as in Google Antigravity.
In short, if you pay for Google AI Pro or Ultra, or you are a developer or enterprise customer on the paid API tier, you can use Neon now. If you use the free Gemini app, you cannot — you will keep getting the standard models, and Neon will simply not appear in your dropdown. That is a deliberate choice by Google: Neon is a premium, work-focused model, and it is priced and gated accordingly.
To try it, paid subscribers can open the Gemini app, tap the model dropdown, and select Gemini 4 Neon. Developers can find it in Google AI Studio under the paid API tier. Enterprise users can ask their Google Cloud account team, and cybersecurity defenders can request the cyber version through the Fairwind Programme.
What happens next
Neon’s launch tells us where Google thinks the AI race is going. The first wave of the Gemini 4 family was about raw capability — Carbon for the hardest problems, Hydrogen for speed. Neon is about economics: the same frontier-level intelligence, packaged for the jobs people actually run every day, at a price and speed that make it practical to use constantly rather than sparingly.
The next test is independent verification. Google’s benchmarks are impressive, but they are Google’s numbers. Over the coming weeks, developers will put Neon through real codebases, researchers will test it on real papers, and security teams will check whether that 76 per cent secure-code score survives contact with real vulnerabilities. If the independent results match Google’s claims, Neon could become the default workhorse model for a large part of the AI industry. If they do not, it will be remembered as another strong launch with soft follow-through.
For now, the facts are these: Neon is live, it is fast, it is cheaper than Carbon, it tops several coding and automation benchmarks, and it is only for paying users. If you build software, analyse data, or defend systems for a living, it is worth a look. If you use the free Gemini app, you will have to wait — or pay — to find out what it can do.
Frequently Asked Questions
What is Gemini 4 Neon?
Gemini 4 Neon is Google’s newest AI model, launched on November 18, 2026. It is designed for fast end-to-end completion of complex, multi-step tasks such as software engineering, data analysis, scientific research and cybersecurity, with a 1 million-token context window that can handle spreadsheets, PDFs, charts, audio and video.
How much does Gemini 4 Neon cost?
Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens through the Gemini API. Cached input tokens cost $0.15 per million, a 95 per cent discount. Google says it runs about three times faster than Gemini 4 Carbon while costing around 60 per cent less.
How good is Gemini 4 Neon in benchmarks?
Neon ranks third overall with a score of 61 on the Artificial Analysis Intelligence Index, second on the Vals Index, and first on the Artificial Analysis Coding Agent Index with a score of 58.6. It also scores 65.2 per cent on Terminal-Bench 4.1 and 46.2 per cent on FrontierCode 2.0, and ranks first on Zapier’s AutomationBench and Andon Labs’ Vending-Bench 2.
Is Gemini 4 Neon available on the free Gemini plan?
No. Gemini 4 Neon is not available on the free Gemini plan. It is available to paid Google AI Pro and Ultra subscribers through the Gemini app and Google Antigravity, to developers through the paid tier of the Gemini API in Google AI Studio, and to enterprises through Gemini for Business.
What is Gemini 4 Neon for cyber?
Gemini 4 Neon for cyber is a special version built to help security teams write, test and repair secure code. It scores 76 per cent on the hard tier of AutoBaxBench, 90.5 per cent on SEC-bench Pro for finding vulnerabilities and 70.7 per cent on CWE-bench v2 for patching them. It is rolling out to trusted defenders through Google’s Fairwind Programme.
Related articles on TechInKey
- Google Launches Gemini 4 Hydrogen — a Fast AI Model Built for Coding and Agentic Workflows
- Gemini Tasks Is Now Free: Google’s AI Agent That Calls, Books and Pays
- Gemini Finally Lets You Keep Your Chat History Without Feeding AI Training
Sources: Financial Express (November 19, 2026); The Indian Express (November 19, 2026); Geeky Gadgets (November 19, 2026).

