Technology•

Google's Gemini 4 Neon Scores 61 on a Closely Watched Intelligence Index, Trailing Only OpenAI's GPT-6.2 Astra and Anthropic's Claude Fable 5.2

Google DeepMind's Gemini 4 Neon, released November 18, ranks third on the Artificial Analysis Intelligence Index and fourth on the Vals RSI Index, with a 1 million token context window.

Written By :MINRK
Google's Gemini 4 Neon Scores 61 on a Closely Watched Intelligence Index, Trailing Only OpenAI's GPT-6.2 Astra and Anthropic's Claude Fable 5.2

Photo by BoliviaInteligente on Unsplash

1. What is Gemini 4 Neon?

Gemini 4 Neon is the artificial intelligence model Google DeepMind released on November 18, 2026, the fourth entry in its Gemini 4 lineup. It scores 61 on the Artificial Analysis Intelligence Index, placing third overall behind OpenAI's GPT-6.2 Astra and Anthropic's Claude Fable 5.2, and ranks fourth on the Vals RSI Index, 3.8 points ahead of Gemini 4 Argon.

In practical terms, Neon is a general-purpose model that accepts text, images, video and speech, can hold up to 1 million tokens of material in view at once, and is sold to developers at $3 per million input tokens and $15 per million output tokens. It is positioned inside the Gemini 4 family between two siblings, Gemini 4 Hydrogen and Gemini 4 Carbon.

The launch is notable less for any single specification than for what it says about pace. Google has now put four distinct Gemini 4 models into the market within weeks of one another, an unusually compressed schedule for a company of its size and a sign of how hard the leading developers are pushing to hold or gain ground.

2. Where Neon lands on the leaderboards

Two independent scoreboards frame the release. On the Artificial Analysis Intelligence Index, Neon reaches 61 when run at its highest thinking level. That is good enough for third place overall. The two models above it belong to Google's main rivals: GPT-6.2 Astra from OpenAI and Claude Fable 5.2 from Anthropic.

On the Vals RSI Index, Neon sits fourth overall. The more revealing comparison there is internal. Neon finishes 3.8 points above Gemini 4 Argon, another member of the same family, which gives a rough measure of how much ground Google has covered between two of its own releases in a short span.

The phrase "highest thinking level" deserves a brief explanation. Many current models can be run at different levels of reasoning effort. At a higher setting the model spends more computation working through a problem before it answers, which generally improves results on hard tasks at the cost of time and money. A score achieved at the top setting therefore shows what the model can do when it is given the most room to reason, not necessarily what a user will see on a quick, low-effort query.

Third and fourth place are not first, and that is part of the story too. Google has a model within reach of the top of both tables, but on these two measures the leading positions are still held by its competitors.

3. How Neon fits into the Gemini 4 family

Google has adopted element names for this generation: Hydrogen, Carbon, Neon and Argon. Neon's place is between Hydrogen and Carbon, which makes it one step in a ladder of models and not the top rung by default.

Selling a family instead of a single flagship is now standard practice among the large developers. Different customers want different trade-offs. A company running millions of short customer-service exchanges cares mostly about cost and response time. A research team working through a difficult engineering problem cares mostly about capability and will pay for it. A range of models at different sizes and prices lets one vendor serve both without forcing either to compromise.

A tiered range also changes how the benchmark results should be read. A model placed in the middle of its family is not necessarily meant to post the highest score its maker can achieve. Its job is to offer a particular balance of capability and cost, and a third-place finish on an industry-wide index from a model in that position says something about the strength of the family as a whole.

The speed of the rollout is the more striking feature. Four Gemini 4 models arriving within weeks of each other means Google is filling out its range almost in a single motion, giving developers and enterprise buyers a full menu to choose from early in the generation's life.

4. The 1 million token context window and multimodal input

Beyond the rankings, the specification most likely to matter in daily use is the context window. Neon can take in 1 million tokens at a time. Tokens are the small chunks of text, often a word or part of a word, that a model reads and writes, and the context window is the total amount it can consider in a single pass.

One million tokens is large enough to hold a full novel, an entire software codebase or hours of meeting transcripts in one request. The practical benefit is that users do not have to cut their material into pieces and feed it in a section at a time, a process that tends to lose the connections between distant parts of a document. A model that can see the whole of a contract, a repository or a long series of filings at once can answer questions that depend on how those parts relate.

Neon also accepts multimodal input. It can work with text, images, video and speech, which means a single request can combine, for example, a written question with a chart, a recording or a clip. For businesses, that widens the range of work the model can be pointed at, from reading scanned documents to reviewing recorded calls.

For readers in finance and digital assets, the relevance is direct. Analysing a lengthy regulatory filing, auditing a body of smart contract code or reviewing a long earnings call are all tasks that depend on being able to hold a great deal of material in view at once.

5. Gemini 4 Neon price: $3 in, $15 out

Google has set the price of Neon at $3 per million input tokens and $15 per million output tokens. Input tokens are what the user sends to the model, including any documents attached to a request. Output tokens are what the model writes back. Output is priced higher across the industry because generating text demands more computation than reading it.

The pricing also includes a 95% discount for caching. Caching allows a developer to store material the model has already processed, such as a long set of instructions or a reference document, and reuse it across many requests instead of paying full price to resend it each time. For applications that repeatedly consult the same large body of text, a discount of that size can change the economics considerably.

That matters most in combination with the large context window. Filling a 1 million token window on every request would be expensive at the standard input rate. If much of that material is the same from one request to the next, caching lets a developer take advantage of the window's size without paying the full cost each time.

Price is now as much a competitive lever as raw capability. A model that ranks third on an intelligence index can still win business from the models ranked above it if it delivers comparable results on a given task for less money, which is why buyers increasingly read benchmark tables and price lists side by side.

6. How much weight should the rankings carry?

Benchmark rankings are a useful guide, but they are not a complete account of how a model will perform for a particular user. Each index compresses performance across a chosen set of tests into a single figure, and different indices choose different tests. That is why the same model can sit third on one table and fourth on another without any contradiction.

Small gaps should be read with caution. A difference of a few points between two models may or may not show up in real work, depending on whether the tasks a customer cares about resemble the tasks the benchmark measures. The 3.8-point lead Neon holds over Argon on the Vals RSI Index is a meaningful internal signal for Google, but it does not by itself tell a developer which model will be better for a given job.

The sensible approach for a prospective buyer is to treat the published tables as a shortlist and then test the candidates on real workloads. A team that mostly writes code, one that summarises legal documents and one that answers customer questions will each weigh accuracy, speed, cost and reliability differently, and a model that leads on an aggregate index is not guaranteed to lead on any one of those jobs.

The broader pattern is clearer than any single number. On the Artificial Analysis table, the top three places are occupied by models from three different companies: OpenAI, Anthropic and Google. No single developer has pulled decisively away, and each new release reshuffles the order by a small margin.

7. What the release says about the AI race

Shipping four models in a matter of weeks is a statement of intent. It shows Google using its scale in research and computing to iterate quickly and to cover as many price and performance points as it can, instead of staking its position on one headline model.

For investors, the contest among Google, OpenAI and Anthropic has become one of the central themes in technology markets. Spending on chips, data centres and energy all flows from the expectation that these models will keep improving and that demand for them will keep growing. Each release that lands near the top of the tables reinforces the view that the race remains open, and that the companies involved will need to keep investing heavily to stay in it.

For developers and businesses, the effect is more immediate. More capable models at published prices, with larger context windows and broader input types, expand what can be built and lower the cost of building it. Neon adds another option near the top of the market, and the speed at which Google has assembled the Gemini 4 family suggests that the menu of choices will keep changing quickly.

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