Technology•

Google DeepMind Launches Gemini 4 Neon at $3 per Million Input Tokens, a Mid-Tier Model Aimed at Carbon's Workloads

Google DeepMind released Gemini 4 Neon on November 18, pricing it at $3 and $15 per million tokens and positioning it between the flagship Carbon and the lower-cost Hydrogen.

Written By :MINRK
Google DeepMind Launches Gemini 4 Neon at $3 per Million Input Tokens, a Mid-Tier Model Aimed at Carbon's Workloads

Photo by Immo Wegmann on Unsplash

1. Gemini 4 Neon arrives as a lower-cost option

Google DeepMind released Gemini 4 Neon on November 18, 2026, adding a mid-tier model to its Gemini 4 family that is pitched mainly on cost. Neon is priced at $3 per million input tokens and $15 per million output tokens, carries a 1-million-token context window and is described as performing close to the flagship Gemini 4 Carbon.

The release gives developers and businesses a third point on Google's price and performance curve. Until now, the practical choice within the Gemini 4 line was between a high-performing but slower flagship and a faster, cheaper sibling. Neon is intended for customers who found the first too expensive to run at volume and the second not capable enough for their hardest tasks.

Google DeepMind is the artificial intelligence research division of Google, responsible for the Gemini models that power the company's chatbot, its developer platform and AI features across its products. Each Gemini generation is released as a family of models at different sizes and prices, so that customers can match the model to the difficulty and volume of the work.

Cost is the headline feature because of how modern AI systems are used. A single question to a chatbot is cheap at almost any price. A software agent that calls a model hundreds or thousands of times to complete one job is not, and that is the kind of workload Google is courting with this launch.

2. What is Gemini 4 Neon?

Gemini 4 Neon is the fourth model in Google DeepMind's Gemini 4 series, designed to sit between two siblings with very different profiles. It is a general-purpose large language model, offered at launch through Google's developer and consumer products, that aims to deliver performance near the top of the range at a lower cost per task.

The models around it define its role. Gemini 4 Carbon is the high-performance option, but it is slower. Gemini 4 Hydrogen is quicker and less expensive. Neon targets the space between them, promising better speed and cost efficiency than Carbon while staying close to it on quality.

Google is directing the model at demanding professional work. The target uses include complex software engineering, knowledge work and scientific applications. What those categories share is repetition: they tend to involve many model calls strung together, so that a modest saving on each call compounds across an entire project.

The naming also reflects how quickly the family has grown. Neon arrived during a rapid expansion of the Gemini 4 lineup in late 2026, by which point DeepMind had already launched several models under that banner, Argon and Carbon among them.

3. Pricing, context window and thinking levels

Published rates put Neon at $3 for every million input tokens and $15 for every million output tokens. Tokens are the units of text that language models read and write, each roughly a fragment of a word, and providers bill separately for the text sent to a model and the text it generates. Output is typically the more expensive side, as it is here.

The model's context window is 1 million tokens. The context window is the amount of material a model can consider at once, covering the instructions, any documents or code supplied and the conversation so far. A window of that size allows very large inputs, such as lengthy contracts, sets of financial filings or substantial portions of a codebase, to be handled in a single request.

Neon also supports different thinking levels, with a high setting as the default. A thinking level governs how much reasoning the model carries out before it responds. More thinking generally produces better answers, and it also raises the cost of each task, because the reasoning itself consumes tokens that are billed.

That trade-off is why the cost of a task is not the same thing as the listed price per token. Two models with identical token rates can cost different amounts to finish the same job if one of them reasons at greater length or needs more attempts. For buyers, the meaningful comparison is what it costs to get a piece of work done to an acceptable standard, and Google is presenting Neon as the less expensive route than Carbon on that measure.

4. How Neon performs on benchmarks

Independent scoring places the new model near the top of the field. On the Artificial Analysis Intelligence Index, Neon ranks third with a score of 61 when run at its highest thinking level.

That index is a composite measure that combines results across a range of evaluations into a single figure, which makes it a common shorthand for comparing models from different developers. A third-place position for a model that is not its maker's flagship is the core of Google's argument: buyers give up little in capability by stepping down a tier.

The areas in which Neon is described as strongest are analytical. Legal research, finance and accounting are singled out, along with other work that depends on careful reading and reasoning over documents.

It is also presented as capable in agentic programming tasks, specifically multi-file coding and tool execution. Multi-file coding means making coordinated changes across several files in a software project instead of editing a single snippet. Tool execution refers to the model calling outside software, such as running code or querying a system, as part of completing a task. Both are central to the AI agents that companies are now building to carry out multi-step work with limited supervision.

5. Where the model is available

Access opened on launch day, with no waitlist. Developers can reach Neon through the Gemini API, the programming interface that lets companies build Google's models into their own products and internal tools.

It is also offered in Google Antigravity to paid subscribers, and in the Gemini app, the company's consumer-facing assistant. Releasing across all three at once puts the model in front of software teams, professional users and the general public simultaneously, without the staged rollouts that have accompanied some earlier AI launches.

Immediate availability matters for a model sold on economics. Teams that want to know whether Neon can replace a more expensive model in their own pipelines can begin measuring straight away, using the same interfaces through which they already reach Carbon and Hydrogen. Because all three belong to one family and are served through the same products, moving a workload from one to another is largely a matter of changing which model a request is sent to and then checking the results.

6. Why a cheaper mid-tier model matters

Demand for this kind of product has been building as artificial intelligence moves out of the experimental phase. Many companies spent the past few years running pilots, where the volume of requests was small and the bill was an afterthought. As those projects move into production, the cost of each task is multiplied across every customer, document and transaction the system touches.

Neon is aimed squarely at high-frequency operational tasks of that sort. In those settings, a model that is slightly less capable than the best available can still be the right choice if it clears the quality bar and costs less each time it runs. A model that is close to the flagship on quality and cheaper to operate removes much of the reason to pay for the flagship on routine work.

The same logic applies to speed. Carbon's slower responses are a real constraint for agents that must take many steps in sequence, since delays accumulate just as costs do. A quicker model shortens the time an automated workflow takes to finish, which can matter as much to a business as the invoice.

7. Who has the most reason to switch

Existing Carbon customers are the most obvious audience. If Neon matches or comes close to Carbon on the quality of its output while costing less to run, those users have a strong case for at least testing whether their workloads can be moved to it.

Legal, finance and accounting teams have the clearest incentive of all. Those are the fields in which the model is said to perform best, and firms in those sectors process large volumes of documents, the kind of repetitive, high-volume analytical work where a lower cost per task adds up quickly.

Software teams building agents form a second group. Multi-file coding and tool use are the capabilities such systems lean on most heavily, and they are also among the most call-intensive uses of a model. A lower-cost option that handles them well changes the economics of running coding agents at scale.

8. What to watch next

The main open question is how large the saving proves to be in practice. The advantage over Carbon is not a single fixed number. It varies with the comparison being made, including the thinking level at which each model is run, so the benefit a given customer sees will depend on the configuration and the type of work involved.

That makes real-world testing the next step for prospective users. Companies weighing a move will want to run their own tasks through both models and compare the total cost and the quality of the results, instead of relying on list prices or a single benchmark figure.

For Google DeepMind, the launch rounds out a lineup that now spans a flagship, a budget tier and a model in between, alongside the other Gemini 4 releases of recent months. How customers distribute their work across those tiers will show whether a near-flagship model at a lower cost draws usage away from Carbon or brings in new workloads that were previously too expensive to automate.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY