Google has introduced Gemini 4 Neon, a new artificial intelligence model designed to combine advanced reasoning capabilities with faster execution and lower operating costs. Announced on November 18, 2026, the model targets complex software engineering, research, analytical work and enterprise automation, positioning itself between everyday AI assistants and Google’s most demanding frontier models.
Google says Gemini 4 Neon matched or exceeded Gemini 4 Carbon on overall quality in its early evaluations while operating at three times the speed and approximately 60% of the cost. The model comes with a one-million-token context window and is rolling out across the Gemini app, Google AI Studio, Gemini API, Gemini for Business and Google Antigravity, with availability varying by subscription and platform.
Key Takeaways
- Google launched Gemini 4 Neon on November 18, 2026.
- The model is designed for complex coding, research, reasoning and enterprise workflows.
- Google says Neon delivers comparable or better quality than Gemini 4 Carbon in early evaluations, at three times the speed and 60% of the cost.
- API pricing starts at $3 per million input tokens and $15 per million output tokens.
- It supports a one-million-token context window.
- The model is rolling out across Google’s AI products, including Gemini, AI Studio and Antigravity.
- Google says Neon performs strongly in agentic coding, scientific reasoning and professional knowledge-work benchmarks.
What Is Gemini 4 Neon?
Gemini 4 Neon is Google’s latest addition to its Gemini 4 family of artificial intelligence models. It is designed to handle demanding tasks that require more than simple question answering, including multi-step software development, complex research, spreadsheet analysis and automated business workflows.
The model is adapted from Gemini 4 Carbon, Google’s larger model designed for its most difficult and extended tasks. Instead of positioning Neon as a completely separate flagship system, Google is presenting it as a more efficient option that brings advanced reasoning capabilities to workflows where speed and operating costs matter.
This distinction is important for businesses and developers. Larger AI models can deliver sophisticated results, but their computational requirements and response times can make them expensive to use repeatedly. A model that approaches their performance while responding faster could make advanced AI agents more practical for everyday business operations.
Google describes Neon as a model for complex, high-frequency workflows. It is intended to work independently or serve as an orchestrator that coordinates specialised AI agents.
Gemini 4 Neon Features and Capabilities
1. Advanced Coding and Software Engineering
Software development is one of Gemini 4 Neon’s central use cases. Google says the model is built to handle long-running coding workflows, including debugging, refactoring and working across multiple files in large repositories.
Rather than simply generating isolated code snippets, Neon is designed to understand broader software projects, investigate problems, make targeted changes and work through multiple steps.
Google reported a 65.2% score on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0. It also reported a 39% score on SWE-sweep using Google’s Antigravity evaluation environment.
These results suggest that Neon is intended to compete in the increasingly important category of AI coding agents, where models must complete practical development tasks rather than merely produce code.
2. One-Million-Token Context Window
Gemini 4 Neon supports a context window of up to one million tokens.
A large context window allows the model to process substantial amounts of information within a single workflow. This can be useful when analysing long reports, reviewing large codebases, comparing documents or combining information from multiple sources.
The model also accepts text, images, audio and video inputs, according to Google’s model documentation.
For businesses, this could reduce the need to divide large projects into numerous smaller prompts. However, a large context window does not guarantee perfect recall or accuracy. Users still need to verify important findings and outputs.
3. Enterprise Research and Knowledge Work
Google is also targeting professional knowledge work, where employees regularly deal with spreadsheets, presentations, PDFs, charts and other information sources.
Gemini 4 Neon is designed to combine information from these different formats and turn it into structured outputs. Potential applications include financial analysis, document research, spreadsheet modelling, reporting and business process automation.
Google reported strong results on several knowledge-work evaluations, including a 56.3% score on Zapier’s AutomationBench and 75.8% on Agents’ Last Exam v1.1.
The company also highlighted performance in finance, accounting and spreadsheet-related benchmarks.
4. Scientific and Mathematical Reasoning
The model is designed for technical research and scientific problem-solving in addition to business applications.
Google reported scores of 84% on RiemannBench, 70.5% on Terminal-Bench Science and 71.3% on HLE-Diamond.
These benchmarks cover different types of mathematical, scientific and advanced reasoning tasks. Strong results could make the model useful for researchers and technical teams that need help analysing complex information.
However, benchmark performance should not be interpreted as proof that an AI model can independently conduct scientific research or replace expert review.
Gemini 4 Neon Performance: What Google’s Benchmarks Show
Google has published results across coding, knowledge work, science, mathematics, reasoning and other categories.
| Benchmark | Gemini 4 Neon score |
|---|---|
| Artificial Analysis Intelligence Index | 61 |
| Vals Index | 70.9% |
| Terminal-Bench 4.1 | 65.2% |
| FrontierCode 2.0 | 46.2% |
| SWE-sweep (Antigravity) | 39.0% |
| AutomationBench | 56.3% |
| Agents’ Last Exam v1.1 | 75.8% |
| RiemannBench | 84.0% |
| Terminal-Bench Science | 70.5% |
| HLE-Diamond | 71.3% |
Source: Google DeepMind’s Gemini 4 Neon announcement and model card. Results are based on specific benchmarks and evaluation settings.
Google also reported that Neon ranked third overall on the Artificial Analysis Intelligence Index, with a score of 61 at its xhigh thinking level. On the Vals Index, the company reported a second-place ranking.
These results are notable, but benchmark rankings do not necessarily predict how a model will perform in every real-world application. Independent testing, practical deployment costs and reliability over extended tasks will remain important measures.
Gemini 4 Neon Pricing: How Much Does It Cost?
Google has announced the following Gemini 4 Neon API pricing:
| Pricing category | Cost |
|---|---|
| Input tokens | $3 per million tokens |
| Cached input tokens | $0.15 per million tokens |
| Output tokens | $15 per million tokens |
| Context window | Up to 1 million tokens |
The cached-input price represents a 95% discount compared with the standard input-token price.
This pricing structure is relevant for developers running AI applications that repeatedly process similar information. Cached inputs can reduce costs when previously supplied context can be reused.
For businesses deploying AI agents at scale, however, the final bill will depend on several factors, including the volume of input and output tokens, the number of model calls, the complexity of tasks and the surrounding infrastructure.
The three-times-faster and 60%-of-the-cost comparison with Gemini 4 Carbon is Google’s own early evaluation claim, not a guarantee that every application will achieve the same savings.
Where Is Gemini 4 Neon Available?
Google is rolling out Gemini 4 Neon across several products.
| Platform | Availability |
|---|---|
| Gemini app | Google AI Pro and Ultra subscribers |
| Google AI Studio | Gemini API and eligible paid access |
| Gemini API | Paid API tier |
| Google Antigravity | Eligible Google AI Pro and Ultra subscribers |
| Gemini for Business | Through the model dropdown and API |
| Android Studio | Developer access through Google’s AI development ecosystem |
Availability and access limits may differ depending on the product and subscription.
For individual users, the most direct route is through the Gemini app’s model selector if their subscription includes Neon. Developers can access the model through Google AI Studio and the Gemini API.
Google Antigravity is particularly relevant because Neon is designed to support extended, tool-heavy development workflows and coordinate specialised agents.
How Gemini 4 Neon Fits Into Google’s Gemini 4 Strategy
Gemini 4 Neon is part of Google’s broader effort to offer different models for different levels of AI work.
Google’s model family includes systems designed for everyday assistance, complex reasoning and more demanding long-running tasks. Neon is positioned as a balance between advanced capability and efficient execution.
The company’s positioning can be summarised as follows:
| Model | Intended positioning |
|---|---|
| Gemini 4 Hydrogen | Fast everyday assistance |
| Gemini 4 Neon | Advanced reasoning with faster, more efficient execution |
| Gemini 4 Carbon | More demanding and complex long-horizon work |
| Gemini 4 Argon | Part of Google’s broader Gemini 4 model lineup |
This strategy reflects a growing shift in the AI industry. Companies are increasingly offering multiple models rather than relying on a single system for every task.
A lightweight model may be sufficient for routine questions, while a more capable model may be necessary for software engineering, research or complex business analysis. The challenge is to select the right model without unnecessarily increasing costs or slowing down applications.
Neon is Google’s attempt to address that middle ground.
What Does Gemini 4 Neon Mean for Businesses and Developers?
The launch could have implications for companies that want to move AI from experimentation into regular business operations.
Lower Costs for AI Agents
AI agents often require multiple model calls to complete a task. They may need to search documents, analyse information, use tools, check results and revise their work.
If Neon can deliver strong results at a lower cost, companies could potentially run more agent-based workflows within existing budgets.
Faster Software Development
Developers could use Neon for debugging, code reviews, repository exploration and multi-step engineering tasks. Faster execution may also make interactive coding assistants more responsive.
More Practical Enterprise Automation
Businesses could apply the model to financial reporting, document processing, spreadsheet analysis and internal research.
However, organisations will still need governance systems, human review, access controls and reliable testing before using AI-generated outputs in important decisions.
Greater Competition in the AI Model Market
Google’s launch adds another option for businesses comparing advanced AI models.
The competition is no longer based only on which model achieves the highest benchmark score. Speed, cost, reliability, context length, tool use and deployment flexibility are becoming equally important.
Limitations and Safety Considerations
Despite its performance claims, Gemini 4 Neon is not free from the limitations associated with generative AI.
Google’s model card acknowledges that the model can produce hallucinations. It also identifies ongoing work on jailbreak resistance and indirect prompt-injection attacks.
The model’s documented knowledge cutoff is March 2026, meaning users should not assume that its underlying training knowledge includes developments after that date. Connected search and other tools can help retrieve current information, but retrieved information still needs verification.
Google says Neon retains protections from its Gemini 4 safety and security stack, including safeguards against certain cybersecurity and biological misuse risks. The company has also reported improvements in resisting malicious instructions embedded in documents, websites and other external material.
For enterprise deployments, these safeguards are important but should not replace independent security testing and appropriate human oversight.
The Bigger Picture
The launch of Gemini 4 Neon highlights a major change in the economics of artificial intelligence. As AI models become more capable, the next competitive advantage may come from delivering that intelligence at a speed and cost that makes large-scale deployment commercially practical.
For Google, Neon offers a way to bring advanced reasoning into more frequent workflows without requiring businesses to use their most expensive model for every task. This could make AI agents more attractive for software development, research, finance and enterprise automation.
The launch also demonstrates why model benchmarks are only one part of the competition. Companies need to evaluate how models perform in their own environments, how much they cost to operate and whether they can complete tasks reliably over extended periods.
Looking Ahead
The next important test for Gemini 4 Neon will be its performance in real-world deployments. Developers and businesses will need to compare its speed, accuracy, token consumption and reliability against competing models across their actual workloads. Independent evaluations will help determine whether Google’s reported performance advantages translate into meaningful improvements outside its own testing environments.
For the broader AI market, the launch reinforces the importance of efficient frontier models. As businesses move from simple chatbots towards AI agents capable of completing multi-step tasks, models that combine strong reasoning with lower costs and faster execution could become increasingly important. Gemini 4 Neon is Google’s latest attempt to compete in that market.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



