Google unveiled its new Gemini 4 Neon AI model. [Photo: Google]

[DigitalToday reporter Jae-won Choi (최재원)] Google has unveiled Gemini 4 Neon, a new AI model that combines high-performance reasoning, fast processing speed and cost efficiency. It is aimed at offering performance comparable to top-tier AI models for complex tasks such as coding, scientific research and enterprise work automation.

Gigazine, a Japanese media outlet, reported on Nov. 19 that Google officially launched Gemini 4 Neon a day earlier, on Nov. 18. It was developed based on the top-tier Gemini 4 Carbon model, and sits between Carbon, which handles complex reasoning, and Gemini 4 Hydrogen, which is specialised for everyday tasks.

Neon supports text, image, audio and video inputs and offers a context window of up to 1 million tokens. Under typical synchronous processing it can output up to 64,000 tokens, and up to 1 million tokens for background long-form generation.

It also ranked highly in performance evaluations. On AI evaluation firm Artificial Analysis' overall intelligence index, it scored 61 under the 'xhigh' benchmark for the highest reasoning level, placing third overall. It outscored Claude Opus 5.5, which scored 59, and GPT-6.2 Sol, which scored 57.

Its coding performance was more pronounced. It ranked first on a coding agent index with a score of 58.6. It achieved 65.2 percent on Terminal-Bench 4.1 and 46.2 percent on FrontierCode 2.0.

It scored 56.3 percent on AutomationBench, an evaluation of work automation, and 70.5 percent on Terminal-Bench Science in the science category. In some cases, however, competing models performed better depending on the evaluation items and execution environment.

Google also highlighted cost efficiency. It said the cost per task at the base reasoning level is 32 percent lower than Carbon, and at the highest level is more than 36 percent cheaper than competing top-tier models.

API pricing is $3 per 1 million input tokens, about 4,000 won, and $15 per 1 million output tokens, about 20,000 won. Cached inputs are offered at $0.15, about 200 won, reflecting a 95 percent discount.

Google also announced Gemini 4 Neon for cyber, a cybersecurity-focused model. It scored 90.5 percent on SEC-Bench Pro, a vulnerability detection evaluation, and 70.7 percent on CWE-Bench v2, a vulnerability-fixing evaluation.

Google said internal tests showed its hallucination rate was the lowest among existing Gemini models. It also recorded a 0 percent attack success rate in an evaluation of defences against prompt injection under certain conditions, but that does not mean it blocks all security threats.

Neon will be rolled out sequentially through Google AI Studio, the Gemini API and enterprise services. Google AI Pro and Ultra subscribers can also use it in the Gemini app and Antigravity. The security-focused model will be provided on a limited basis to verified experts.

Keyword

#Google #Gemini 4 Neon #Artificial Analysis #Gemini API #Gemini AI Studio
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.