Google launches Gemini 4 Neon to balance high performance and cost
The new model matches the performance of Google's top-tier Carbon model while offering faster processing and lower long-term costs for enterprise users.
The short version
- Neon matches Carbon's performance in several benchmarks, including intelligence and coding tasks.
- The model will become 25% cheaper than Carbon per token once introductory pricing ends on November 30.
- Neon supports a context window of one million tokens and processes text, images, audio, and video.
- Google positions Neon for complex programming and analytical reasoning, while Carbon remains for long-term projects.
Why it matters
For businesses, Neon provides a way to access high-level reasoning without the premium price tag of the most powerful models. It signals a shift toward 'efficient' high-performance AI that can handle complex tasks like coding and analysis more economically.
A new middle ground for AI
Google DeepMind has unveiled Gemini 4 Neon, the fourth model in the Gemini-4 family released within a seven-week span. This new addition is specifically designed to fill the gap between high-end, heavy-duty models and those built for simple, routine tasks. While Google offers Hydrogen for everyday tasks, Neon is intended for more complex work like analytical reasoning and professional enterprise use.
The model is derived from Google's flagship Gemini 4 Carbon. Despite being optimized for efficiency, Neon manages to reach the same performance levels as its larger sibling in many key areas. In some specific benchmarks, Neon even slightly outperforms Carbon, making it a formidable competitor for high-level reasoning tasks.
Google is organizing its lineup into a three-part structure to help users choose the right tool for their specific needs. Neon will serve as the primary model in the Gemini app, while Carbon is reserved for the most demanding long-term projects. The previous model, Argon, which was also aimed at knowledge work, is being phased out of this new three-part structure.
Benchmark performance and coding
According to data from Artificial Analysis, Neon's highest thinking level, known as Xhigh, scored 61 points on the Intelligence Index. This puts it in third place globally, trailing only GPT-6.2 Astra and Claude Fable 5.2. Notably, this score is slightly higher than the 60 points achieved by Google's own Carbon model.
The model shows specific strengths in agentic knowledge work, image understanding, and programming. In the Coding Agent Index, Neon scored 59 points, which is comparable to top-tier models from Anthropic. However, it is important to note that these scores were achieved using Google's Antigravity CLI rather than the specific tools used to test its competitors.
In the Vals Index, which measures performance in practical fields like finance, law, and taxes, Neon ranked second. It sits just behind Claude Fable 5.2 and slightly ahead of Carbon. These results suggest that Neon is highly capable of handling professional-grade inquiries that require more than just basic information retrieval.
Efficiency and pricing shifts
One of the most significant aspects of the Neon launch is its cost-to-performance ratio. Currently, Neon costs $3 per million input tokens and $15 per million output tokens via the Gemini API. Because Carbon is currently on an introductory price plan, Neon is actually 50% more expensive than Carbon until the end of November.
The pricing landscape will shift significantly on December 1. When Carbon's introductory rates expire, its prices will rise to $4 and $20 per million tokens respectively. At that point, Neon will become 25% cheaper than Carbon. This transition makes Neon the more economical choice for users who need high-level performance but want to manage their spending.
Testing by Artificial Analysis shows that Neon is also faster than its larger counterpart. At its maximum thinking level, Neon completed tasks in about 21 minutes, while Carbon took 36 minutes. This means Neon requires roughly 40% less time and costs about 17% less than Carbon when comparing their regular prices.
Technical specs and availability
Neon is a multimodal model, meaning it can process various types of data including text, images, audio, and video. It is designed to output text and features a context window of one million tokens. A context window refers to the amount of information the model can keep in its active memory at one time during a conversation.
The model's training knowledge ends in March 2026. It also features four different thinking levels to allow users to scale the complexity of the AI's reasoning based on the task at hand. This flexibility allows it to handle everything from quick questions to deep analytical work within a single framework.
Users can access Neon through the Gemini API, Google AI Studio, and Gemini for Business. While there is no free API access, subscribers to Google AI Pro and Ultra can use it within the Antigravity development environment. In the standard Gemini app, Carbon remains exclusive to the more expensive Ultra tier.
What to watch
- How businesses adopt Neon as the primary model for coding tasks.
- Whether other AI providers release 'efficiency' models that match Neon's speed-to-cost ratio.
Try it
Check the Gemini API pricing page to see how the costs for Neon and Carbon will change starting December 1.
Sources and links
- OfficialVals Index vals.ai
- Officialsagt Google blog.google
- Reportingheise.de: the article we sourced
First reported by heise.de. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.