Gemini 4 Neon: Google’s Fast, Cheaper Frontier AI Model Explained (2026)

Gemini 4 Neon is Google’s newest frontier AI model, launched on 18 November 2026, and it is built for one job: doing serious work fast, without the serious bill. It sits in the middle of Google’s Gemini 4 family — stronger than the speedy Hydrogen model, cheaper and quicker than the heavyweight Carbon model — and it arrives as every major AI lab cuts prices to win professional work.
The headline number is the price. Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, with cached input at just $0.15 per million tokens. That is less than half the $8 per million input tokens that OpenAI charges for GPT-6.2 Astra and Anthropic charges for Claude Fable 5.2.
What Is Gemini 4 Neon?
Gemini 4 Neon is the fourth member of Google’s Gemini 4 family, following Argon in October, Carbon on 10 November and Gemini 4 Hydrogen on 12 November. Google describes it as a frontier-level model built for complex software engineering, knowledge work and science workflows. In plain terms, it is the model you pick when a task is too hard for a fast, cheap model but does not need the slowest, most expensive one.
It comes with a 1 million-token context window, so it can read very large documents, codebases or data sets in one go. It also has four thinking levels — Low, Medium, High and Extra High — so you can choose how much reasoning effort each task gets. A quick question can run on Low and cost almost nothing; a hard coding or research job can run on Extra High for the best result. Google also says Neon records its lowest fabrication rate yet in internal Antigravity testing — important for legal and financial work, where invented facts are useless.
How Much Does Gemini 4 Neon Cost?
The full price list is simple: $3 per million input tokens, $0.15 per million cached input tokens (a 95% discount) and $15 per million output tokens. There is no introductory or promotional price — the rate is flat.
Compare that with the competition. OpenAI’s GPT-6.2 Astra and Anthropic’s Claude Fable 5.2 both charge $8 per million input tokens. Anthropic’s larger flagship, Claude Opus 5.5, charges $4 per million input tokens and $20 per million output tokens — so Neon is 25% cheaper than Opus 5.5 while scoring close to it on several engineering benchmarks.
Inside Google’s own family, Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens. Gemini 4 Carbon, the model for the hardest long-horizon projects, is on a promotional price of $2 per million input tokens until 30 November 2026, after which it rises to $4 and $20. From 1 December, Neon will actually be cheaper than Carbon.
Gemini 4 Neon Benchmarks: Where It Wins
Benchmarks are vendor tests, so they should always be read with care. But Neon has been checked by independent indices too, and the results are strong.
On the Artificial Analysis Intelligence Index, Gemini 4 Neon scores 61 at its highest thinking level, ranking third overall — just behind larger, slower models. On the Vals Index, which measures professional work, it ranks second with 70.9%, and first on both the Legal Research Bench and the Tax Agent Bench. For coding, it ranks first overall on the Artificial Analysis Coding Agent Index with 58.6, and scored 65.2% on Terminal-Bench 4.1.
The business-automation results are just as striking. Gemini 4 Neon ranks first on Zapier’s AutomationBench with 56.3%, and first on Andon Labs’ Vending-Bench 2 — a long-horizon simulation of running a business — with a score of $16,843. It also debuts second globally on the new Agent Arena Index with 1,336 points across 4,200+ real-world agent sessions, and ranks second on steerability — how cleanly it adapts when corrected mid-task.
On science and maths, it is best in class on Terminal-Bench Science (70.5%), RiemannBench (84.0%) and HLE-Diamond (71.3%). One early user quote stands out: Thai Tran, AI product lead at Glean, said Neon “matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost” in early evaluations. Google gives no completion times, so treat that as an early-user report.
Gemini 4 Neon vs Hydrogen, Carbon and the Competition
The Gemini 4 family now has a clear shape. Gemini 4 Hydrogen is the fast, light workhorse for everyday execution — it completes tasks 2.5 times faster than Gemini 3.8 Flash. Gemini 4 Carbon is the model for your hardest, longest projects. Gemini 4 Neon sits between them: Carbon-level quality at Flash speeds, as 9to5Google put it, and cheaper than Carbon once the promotional pricing ends.
Against rivals, the picture is similar. OpenAI’s GPT-6.2 Astra scores 62 on the Artificial Analysis Intelligence Index against Neon’s 61, but costs more than twice as much per input token. Anthropic’s Claude Fable 5.2 scores 65, again at $8 per million input tokens. Google’s head-to-head tests show Neon beating OpenAI’s mid-range GPT-6.2 Sol on business automation, while Sol reads dense charts more accurately. No single model wins everywhere.
This is also a week of big money moving around AI: Anthropic’s $83 IPO debut put a public price on the Claude maker, and every lab is now competing on the cost of professional work, not just on raw capability. Cheaper frontier models mean more teams can afford to use AI for real work.
Where You Can Use Gemini 4 Neon
Gemini 4 Neon is rolling out now in the Gemini API in Google AI Studio, in Gemini for business, in the Gemini app for paid Google AI Pro and Ultra subscribers, and in Google Antigravity for paid Pro and Ultra subscribers. It replaces Argon, Gemini 3.1 Pro is being retired, and Gemini 3.5 Flash-Lite is now labelled “Auto” in the model picker for paid users.
What Gemini 4 Neon Means for You
For most people, the practical meaning is simple: frontier-level AI is getting cheaper and faster, and the best model for your job may no longer be the most expensive one. A small business can now run contract review, tax research, spreadsheet modelling and coding help on a model that ranks first on the relevant professional benchmarks, for about a third of what the same work cost on a top-tier model a month ago.
That does not mean switching everything tomorrow: vendor benchmarks are measured on the vendor’s own tests. But if you pay for AI by the token, Neon is worth a test run this week. The same shift is reaching everyday tools — OpenAI’s ChatGPT Wallet already lets its agents handle purchases with your approval — and cheaper models will push that further.
A Simple Starter Plan
If you want to try Gemini 4 Neon this week, keep it small and measurable:
- Pick one real task — a contract summary, a spreadsheet model or a coding fix — and run it on Neon at the High thinking level.
- Run the same task on the model you use today, and compare quality, time taken and token cost.
- Try the cached-input discount: repeat a task with the same documents attached and watch the input cost fall by 95%.
- Check the output carefully before you trust it, especially for numbers, dates and legal wording.
- If Neon matches your current model on quality, switch the routine jobs to Neon and keep the hardest work on the top model.
Frequently Asked Questions
Is Gemini 4 Neon better than GPT-6.2 Astra?
It depends on the job. Astra scores slightly higher on the general Artificial Analysis Intelligence Index (62 vs 61), but Neon costs less than half as much per input token and beats Astra on several professional and coding-agent benchmarks. For most business work, Neon is the better value.
How much does Gemini 4 Neon cost?
$3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, with a 1 million-token context window. There is no promotional price, so the rate stays the same after 30 November.
Where is Gemini 4 Neon available?
In the Gemini API in Google AI Studio, in Gemini for business, in the Gemini app for Google AI Pro and Ultra subscribers, and in Google Antigravity for paid Pro and Ultra subscribers. It replaces Argon in the Gemini 4 family.
The Bottom Line
Gemini 4 Neon is not the biggest model Google has shipped. It is something more useful: a frontier-level model priced for everyday professional work. It ranks third overall on the independent Artificial Analysis Intelligence Index, first on coding-agent and legal-research benchmarks, and costs less than half of what OpenAI and Anthropic charge for their top models. The AI price war is a race to the bottom on cost and to the top on reliability — and on the evidence so far, Neon is a strong bet.
Sources: Google’s launch announcement (18 Nov 2026), PYMNTS (19 Nov 2026), VentureBeat, 9to5Google.






