Google's New Gemini 4 Neon Costs More Per Token Than Its Flagship
Neon runs $3 per million input tokens, while flagship Carbon sells at an introductory $2 that's set to double.

Google's new Gemini 4 Neon, pitched as the thrifty alternative to its flagship AI model, costs more per token than that flagship until Dec. 1. Google DeepMind announced Neon on Wednesday at $3 per million input tokens and $15 per million output tokens, the meter developers get billed on. Gemini 4 Carbon, the company's top model, is selling for an introductory $2 and $10.
So Google isn't selling Neon on its rate. It's selling it on how much of the meter a job runs up. By Google DeepMind's own numbers in the Neon announcement, a task at Neon's default "high" thinking setting costs 32% less than the same task on Carbon. On ExploitGym, a test of whether security flaws can actually be exploited, the company says Neon used half the tokens Carbon did.
The math is simple once it's laid out. The bill is the rate times the tokens, so a model that burns fewer can come out ahead at a higher price. Google says Neon cuts down on redundant file reads. Its Carbon announcement describes that model as built to spend more tokens than Argon exploring, reasoning and checking its own work.
Neon is the fourth Gemini 4 model, and the third in nine days. Carbon arrived Nov. 10 for the hardest long-running jobs. Gemini 4 Hydrogen followed Nov. 12 as the fast everyday model at $1.50 and $7.50, and free Gemini app users get it when automatic model selection picks it for them.
Gemini 4 Argon came before all three. In the Carbon post, Google said Argon was still the right pick for most software and knowledge work. Neon's announcement describes the family as Carbon on top, Hydrogen for speed and Neon in between, and Argon shows up only in the caption of a demo.
Logan Kilpatrick wrote in his launch post on X that Neon "feels like the right balance of intelligence, speed, and usability."
Google did publish one yardstick. It says Neon scored 61 at its highest setting on the Artificial Analysis Intelligence Index, third overall and behind larger, slower models. Hydrogen topped out at 54 on the same index, according to Google.
People who've never bought a token get Neon too. It's in the Gemini app's model menu for Google AI Pro and Ultra subscribers, while Carbon went only to Ultra subscribers in the app. Google's sales pitch is mostly errands: digging through billing statements to find out why a utility charge spiked, or checking receipts against each other to push for a refund.
Developers can reach Neon through the Gemini API's paid tier in Google AI Studio and through Gemini for business. Paid Pro and Ultra subscribers also get it in Google Antigravity, the company's coding tool. A security version is going out to vetted defenders in Google's Fairwind Program over the next few days.
The pricing catches up to the pitch on Dec. 1. That's after Carbon's introductory rate ends Nov. 30 and it moves to $4 and $20, which makes Neon, at $3 and $15, the cheaper meter per token. Google says it's already the cheaper one per task.
Source: deepmind.google, retrieved November 19, 2026. Other sources: deepmind.google (2), deepmind.google (3), X.
™
Comments 0