Google Unveils Gemini 4 Neon, Reducing Work Time by 40%


Google has unveiled its artificial intelligence (AI) model, Gemini 4 Neon. Positioned between the everyday work model 'Hydrogen' and the high-level task model 'Carbon,' Google emphasized its ability to expedite software development, corporate tasks, and scientific research.
 

On November 18, local time, Google announced on its official blog that Neon’s high setting has a cost per task that is 32% lower than Carbon. An independent evaluation agency, Artificial Analysis, reported that Neon scored 61 points in AI intelligence at the highest setting (xhigh), placing it third, one point higher than Carbon (high). The cost per task is $6.44 for Neon, compared to $7.76 for Carbon at standard rates, making it approximately 17% cheaper.
 

In the same evaluation, the average time taken per task was about 21 minutes for Neon, which is approximately 40% shorter than Carbon's roughly 36 minutes. Neon outputs about 110 tokens per second, while Carbon outputs about 57 tokens.
 

The API usage fee is $3 per million tokens for input and $15 per million tokens for output. Carbon has promotional rates (input $2, output $10) until November 30, making Neon more expensive during this period. Starting December 1, Carbon's rates will increase to $4 for input and $20 for output, making Neon 25% cheaper. The previous 17% comparison is also based on standard rates.
 

In evaluations released by Google, Neon ranked second in the enterprise task assessment 'Vals Index' with a score of 70.9%. It ranked first in legal research and tax task evaluations, and first in the task automation assessment 'Automation Bench' with a score of 56.3%. In the scientific field, it recorded scores of 70.5% in 'Terminal Bench Science' and 84.0% in 'Riemann Bench.'
 

The 'Gemini 4 Neon for Cyber' model, specialized for cybersecurity tasks, will be sequentially provided to defense experts approved under Google’s 'Fairwind Program' within a few days. The previously unveiled 'Gemini 4 Argon' is also available only to partners in this program.
 

Neon can process up to 1 million tokens at once and adjusts inference levels across four stages. In the Gemini app, Google AI Pro and Ultra subscribers can access it, while paid subscribers can use it on the development platform Antigravity. The developer API is available through the paid tier of Google AI Studio.





* This article has been translated by AI.