---
title: "Google Raised the Price, Then Sold You the Middle"
date: 2026-11-19
category: Foundation Models
site: NeuroAI
canonical: https://neuroai.site/a/na-model-google-price-move-1119
language: en
---

# Google Raised the Price, Then Sold You the Middle

> On December 1, the price per million input tokens goes from $2 to $4. Output goes from $10 to $20.

On December 1, the price per million input tokens goes from $2 to $4. Output goes from $10 to $20.

Carbon and Argon are the models that move. In the same month Neon arrives at $3 and $15. One doubles, one launches, thirteen days apart.

## Three tiers on one sheet

From December 1, Google's sheet has three tiers: Hydrogen at $1.5 in and $7.5 out, Neon at $3 and $15, Carbon and Argon at $4 and $20.

No overlap. Hydrogen covers daily work, Neon the mainstream, Carbon the hardest long-horizon projects. Users need not know the architecture, only how many runs a day they do and how much a mistake costs.

Output prices run 7.5, 15, 20: the first two tiers double, the third adds 33%. Input prices 1.5, 3, 4 follow the same shape. These are not isolated numbers but a staircase you can read at a glance.

A staircase does not push you to the top. It makes you hesitate between adjacent tiers, where the middle usually wins: stronger than the one below, cheaper than the one above, just enough. The sheet has no recommended slot; the spread is the recommended slot.

## The middle price was engineered

Before the increase, the sheet was a mess. Carbon at $2 and $10 was a third cheaper than Neon at $3 and $15. For weeks the flagship was cheaper than the middle model while sitting a tier above it, and no user had a reason to choose Neon.

After December 1 the order flips. Carbon at $4 and $20 is a third more expensive. Same Neon, same sheet, thirteen days apart, from most expensive to middle. What changed was not Neon but the number beside it.

The filings do not address existing contracts; new prices apply from December 1 to renewals and new business. Enterprise buyers rebuild a budget, and a developer's long task may cost twice as much. By the calendar Neon shipped first; by the bill most users see both prices after December.

## The arithmetic on one long task

Take one million input and 200,000 output tokens.

Hydrogen: 1.5 plus 1.5, $3. Neon: 3 plus 3, $6. Carbon before the increase: 2 plus 2, $4. Carbon after: 4 plus 4, $8.

Same work. Neon is 50% more expensive than the old Carbon and 25% cheaper than the new one. Neon's number never moved; raising Carbon alone turned it into the middle tier. The arithmetic is mine; prices are Google's. No subsidy was paid: users walk to the middle on their own, pushed by the spread.

Caching takes another layer off. A hit costs $0.15, 5% of the input price; a million repeated tokens drop from $3 to $0.15, a 95% saving. The precondition is a stable prefix: system prompt, codebase or contract body first, questions after. The condition for saving is also the manual for use.

A one-million-token codebase read three times a day costs $9 daily, $180 across twenty working days. Cached, reads two and three cost $0.15 each: $3.30 a day, $66 a month. The long-context bill is decided by repetition, not unit price.

## Cost per task is the real number

Google's own measure is cost per task: Neon at the default high setting is 32% below Carbon, 50% to 80% below top rivals, and more than 36% below rivals' top configuration at xhigh. Price per token 50% higher, cost per task 32% lower. Neon uses fewer tokens for the job. Buyers should read the second number; sellers quote the first.

Neither number is your bill. Google has not published the task set behind its figure; it could be the hundred tasks it sees most, or a hundred it selected. 32% indicates direction, not your invoice. Run ten of your own tasks before pushing for a discount.

Argon's installed base is a separate ledger. They paid 2 and 10; from December 1 they pay double. Neon means 3 and 15: 25% below the new price, still 50% above the old. Whether to migrate depends on whether 32% per task covers a 50% rise in unit price. Only usage mix decides.

## Pricing changed coordinates

Models were priced by capability, higher score at a higher price. Now they are priced by task cost: same job, cheaper wins. When the two conflict, users vote with tasks.

Google (NASDAQ: GOOGL) has not explained the December 1 date; the filings give no reason. Three facts are visible: adjacent dates, exact spreads, tiers that fill every gap. No explanation, only a timetable.

The increase also changes how work is assigned. A team used to standardize on the most expensive model to avoid arguments. Now the spread is wide, so who uses which tier goes into process, and procurement shifts from buying a model to buying a sheet. That suits Google: finer segmentation makes switching harder.

Three thresholds: step up and pay double, step down and accept less capability. Most people stop in the middle, which must be both cheap and strong. Neon's 61 sits on that line: 33% more buys capability, one tier down costs it.

A hundred engineers, one long task each per day: $600 a day on Neon, $800 on the new Carbon. Over twenty working days, $4,000 a month. Not enough to buy GPUs, but enough to make procurement reread the sheet.

Open question: with the middle tier 32% cheaper per task, why does the flagship still charge 4 and 20? The answer may be in tasks where one mistake is ruinous, or in the account that says expensive is reliable. Either way, the flagship's pricing power is no longer decided by its score.

The price sheet has no footnotes. It only orders. Users cannot read a parameter table, but they can read the gap between 3 and 15, and 4 and 20.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-model-google-price-move-1119
Free to quote with attribution and a link to the original.
