The Gemini 4 Lineup Isn’t a Simple Ranking
You might expect Google's three Gemini 4 models to form a tidy ladder: a quick, inexpensive option at the bottom, a balanced one in the middle, and the smartest at the top.
Then you look at the numbers. Hydrogen, the cheaper model built for speed, scores above Argon on the Intelligence Index comparison Google shared at launch. Carbon is Google's strongest reasoning model, yet its price per token is identical to Argon's.
That leaves two fair questions. Why choose Argon over Hydrogen? And if Carbon costs the same, why use Argon at all?
The answers have more to do with how the models work through a problem than with where they sit on a leaderboard.
Hydrogen Makes Sense for Everyday Work
Gemini 4 Hydrogen is built for work that needs to keep moving: writing code, reading documents, using tools, and turning a clear request into a finished result. It can also handle agents—software that carries out a sequence of actions, rather than just returning an answer.
“Fast” can sound like a polite way of saying “less capable.” Hydrogen makes that assumption awkward. In Google's launch comparison, it scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, against Argon's 50. At xhigh, Hydrogen reaches 54.
That gives it a serious claim on everyday work. There’s little reason to dismiss it before trying it simply because another model carries the frontier label.
Its four thinking settings also leave room to adjust. Alongside the default high, there’s low, medium, and xhigh. The last option spends more tokens and takes longer, so moving it up for every request would work against Hydrogen’s appeal. A quick document rewrite and a stubborn coding problem don’t need the same treatment.
Why Argon Still Has a Place
A broad benchmark combines many kinds of questions into one score. It can tell you something useful without settling every choice between two models.
Google still positions Gemini 4 Argon for most advanced software engineering and knowledge work. Its distinction from Carbon is clearer: Argon spends fewer tokens per turn and responds faster, while retaining the ability to work through demanding problems. The Argon guide makes that tradeoff explicit.
Consider the difference between adding a feature to a familiar application and investigating a failure nobody can reproduce. Both involve code, but the second can require much more exploration before there’s even a useful change to make.
Argon has a place in that first kind of demanding, fairly well-understood work. It offers deeper reasoning without Carbon’s more extensive investigation on every turn. That doesn’t establish an automatic win over Hydrogen. For a particular application, Hydrogen may already do everything needed.
So Argon’s role is worth understanding, but its place in your stack has to be earned by the results.
Carbon Is Built to Stay With a Difficult Problem
Gemini 4 Carbon spends more thinking tokens exploring possible explanations and checking its work. Google describes it as the family’s strongest model, aimed at difficult engineering, scientific reasoning, and other work that can take many steps to resolve.
The practical difference is easiest to see in debugging. Producing a plausible patch can be quick. Establishing that the patch fixes the cause, rather than hiding one symptom, can take much longer. That may involve testing several explanations, tracing interactions across components, and checking whether the change breaks something elsewhere.
Those are the kinds of investigation Carbon is designed to sustain. Its developer guide explains the additional effort in terms of exploring hypotheses and verifying edge cases.
There’s a reason to pay for that effort when the problem calls for it. There’s less value in asking for the same depth when the task is already clear. Rewriting a short email usually doesn’t leave much room for an extended investigation to improve the result.
Gemini 4 Pricing: The Same Rate Can Produce a Different Bill
As of November 16, 2026, Google lists these standard Gemini Developer API rates. All figures are USD per million tokens; output pricing includes thinking tokens.
Model | Input | Output |
|---|---|---|
Hydrogen | $1.50 | $7.50 |
Argon | $2.00 | $10.00 |
Carbon | $2.00 | $10.00 |
Argon and Carbon’s introductory rates run through November 30. From December 1, both rise to $4 for input and $20 for output. Hydrogen’s listed rates have no corresponding December increase announced. These are Google’s rates, rather than SeeAPI prices; caching, tools, and other service tiers have their own rules on the pricing page.
The identical Argon and Carbon rates explain only the price of each token. They don’t tell you how many tokens a task will consume. If Carbon spends longer checking alternatives, its bill can be higher even when the final answer is the same length.
The reverse can happen too: an expensive successful attempt may cost less than several cheaper failed ones. The useful comparison is what it costs to get work you can accept.
What the Three Models Have in Common
All three accept text, images, audio, video, and PDFs, and produce text. They share a unified context window of 1,048,576 tokens, with input, thinking, and output drawing from the same allowance. Longer generation is available through the long-decode workflow, beyond the default 65,536-token generation limit.
That means choosing Carbon doesn’t buy a larger context window. The extra value has to come from what it does with the material.
For Gemini 4, Hydrogen is a reasonable first trial for everyday work. Argon deserves a look when that work needs more sustained judgment. Carbon makes sense when finding and checking the answer is itself the difficult part. A complicated project can contain all three kinds of work; you don’t have to give every step to the same model.

