Arab AI
Glowing Google Gemini star logo next to Gemini 4 Neon text surrounded by translucent code panels and neural network waves on a dark background

Google Launches Gemini 4 Neon for Advanced Coding and Reasoning

November 19, 2026
7 minutes

Google has introduced Gemini 4 Neon, a new AI model designed for complex software engineering, analytical reasoning, and enterprise workflows. Announced on November 18, 2026, the model combines advanced reasoning capabilities with faster task execution and lower operating costs than larger models, according to Google.

Gemini 4 Neon is rolling out now.

Sitting between Carbon and Hydrogen, it brings frontier intelligence and fast task completion for complex software engineering, knowledge work, and science. pic.twitter.com/qfOLW06FOm

– Google DeepMind (@GoogleDeepMind) November 18, 2026

Gemini 4 Neon features a 1-million-token context window, supports text, images, audio, and video, and is priced at $3 per million input tokens and $15 per million output tokens through the Gemini API. Google is positioning it between Gemini 4 Hydrogen and Gemini 4 Carbon, targeting developers, businesses, and users who regularly handle demanding tasks that require multiple steps and external tools.

Advertisement

The launch also brings new benchmark results spanning coding, mathematics, scientific reasoning, and business automation.

What Is Gemini 4 Neon?

Gemini 4 Neon is part of Google’s fourth-generation Gemini model family. It targets workloads that require substantial reasoning without the higher operating costs of Google’s largest model.

According to Google DeepMind’s model card, Neon is adapted from Gemini 4 Carbon. It is designed to handle complex coding tasks, analytical work, and production-grade enterprise applications, while maintaining faster execution across frequently repeated workflows.

Advertisement

Its 1-million-token context window allows the model to process large amounts of information in a single session. Developers can feed entire codebases, lengthy reports, and financial documents into a single workflow without splitting files.

Neon also supports four reasoning levels: Low, Medium, High, and Extra High (XHigh). Developers can adjust the level according to the complexity of a task and the amount of computational resources they want to spend.

Key Features and Capabilities

Advanced Coding and Software Engineering

Software development is one of the main areas Google highlights for Gemini 4 Neon. The model can work across multiple files in a code repository, identify bugs, debug interconnected components, and refactor existing code with targeted changes.

It is also designed for long-running agent workflows, where completing a task involves several operations rather than generating a single response. Neon can use external tools, track progress, and continue working through complex instructions.

Inside Google Antigravity, it can operate independently or act as an orchestrator, directing multiple Gemini 4 Hydrogen subagents to handle parallel tasks. That orchestration cuts repetitive steps across large, multi-file projects.

Multimodal Analysis and Enterprise Workflows

Gemini 4 Neon also processes images, audio, and video alongside documents and other business data within its unified context window. Users can analyze spreadsheets, PDFs, charts, and multimedia files together in a single prompt.

For example, a business can use the model to examine financial spreadsheets alongside supporting documents, identify inconsistencies, and produce a structured report. Researchers can also combine information from lengthy documents and visual materials when preparing technical analysis.

Neon performs strongly in finance, accounting, spreadsheet modeling, and complex instruction-following tasks.

Scientific and Mathematical Reasoning

The model also targets research and technical workloads that require multi-step reasoning. Its capabilities cover quantitative analysis, advanced mathematics, and scientific problem-solving.

It achieved strong scores on RiemannBench, Terminal-Bench Science, and Humanity’s Last Exam Diamond, three benchmarks for mathematical and scientific reasoning.

Gemini 4 Neon Benchmarks and Performance

Google published a series of benchmark results comparing Neon with models including Claude Opus 5.5 and GPT-6.2 Sol. The results show competitive performance across several categories, although Neon does not lead every evaluation:

BenchmarkGemini 4 NeonClaude Opus 5.5GPT-6.2 Sol
Artificial Analysis Intelligence Index61––
Vals Index70.9%––
CursorBench 4.165.4%––
Terminal-Bench 4.165.2%65.8%–
FrontierCode 2.046.2%47.3%–
SWE-sweep39.0%––
AutomationBench56.3%––
Agents’ Last Exam v1.175.8%––
RiemannBench84.0%––
Terminal-Bench Science70.5%––
HLE-Diamond71.3%––
Chartography73.5%–75.1%

One notable result is its 56.3% score on Zapier’s AutomationBench-the highest result among the compared models-alongside 84.0% on RiemannBench and 70.5% on Terminal-Bench Science.

In software engineering and chart comprehension, however, the comparisons show a more mixed picture: Claude Opus 5.5 edged ahead on Terminal-Bench 4.1 (65.8% vs. 65.2%) and FrontierCode 2.0 (47.3% vs. 46.2%), while GPT-6.2 Sol led on Chartography (75.1% vs. 73.5%).

Overall, Neon’s strongest gains appear in mathematical reasoning and business automation, while rivals retain slight edges in specific coding and chart benchmarks. Because vendors evaluate models under different test setups, these scores highlight relative strengths rather than a single ranking.

Gemini 4 Neon vs Hydrogen vs Carbon

Google’s Gemini 4 lineup includes models designed for different levels of workload complexity:

  • Gemini 4 Hydrogen: Focuses on everyday assistance and fast responses. It is intended for routine tasks that do not require extensive reasoning.
  • Gemini 4 Neon: Targets advanced coding, analytical reasoning, and repeated multi-step workflows, balancing performance with operating costs.
  • Gemini 4 Carbon: Handles the most demanding projects, including complex engineering work and long-running agent tasks.

The launch also changes the model selection experience in the Gemini app. Neon replaces Argon as the option for deeper analysis, while Carbon remains the choice for coding and the most complex work. Hydrogen continues to handle everyday assistance.

This change applies only to the Gemini app’s model picker; Google has not announced a deprecation of Argon in its API.

Gemini 4 Neon Pricing and Availability

Google has set the following prices for Gemini 4 Neon through its API:

UsagePrice per 1 million tokens
Input tokens$3
Cached input tokens$0.15
Output tokens$15

Cached input costs 95% less than standard input, which reduces expenses for applications that repeatedly use the same instructions or context.

The model is rolling out across several Google products and services:

  • Gemini app: Available to Google AI Pro and Google AI Ultra subscribers.
  • Google AI Studio: Available through Google’s developer platform.
  • Gemini API: Available for developers building applications and AI-powered services.
  • Google Antigravity: Available to eligible Pro and Ultra subscribers.
  • Gemini for Business: Accessible through the business platform’s API and model selection options.
  • Android Studio: Available within supported developer tools.

API usage is billed separately from Google AI Pro and Ultra subscriptions, with rate limits varying by plan.

Neon for Cybersecurity

Google has also introduced a cybersecurity-focused version called Gemini 4 Neon for Cyber. It is designed to help security professionals generate safer code, identify vulnerabilities, verify potential exploits, and patch weaknesses in existing software.

In Google’s published evaluations, the base Gemini 4 Neon model scored 51.26% on WebExploitBench v1.2, while the specialized Gemini 4 Neon for Cyber version scored 90.5% on SEC-bench Pro for vulnerability discovery and 70.7% on CWE-bench v2 for vulnerability identification and remediation.

Google plans to roll out Gemini 4 Neon for Cyber exclusively to trusted defenders through its Fairwind Program.

The company has also highlighted protections against indirect prompt injection, where malicious instructions are hidden inside web pages, emails, documents, or code accessed by an AI agent.

What Neon Means for Google’s AI Strategy

With Gemini 4 Neon, Google is building a tiered model lineup that matches specific workloads with appropriate reasoning depth and operating costs.

For developers, this approach makes advanced AI capabilities more practical for everyday software engineering and automated workflows. Businesses also gain another option for analyzing large collections of documents, processing financial data, and coordinating multi-step tasks.

In real-world deployments, developer adoption against OpenAI and Anthropic will hinge on total cost per task and execution speed-two areas where Neon is positioned to compete directly without forcing teams onto Carbon’s higher pricing tier.

Frequently Asked Questions

Is Gemini 4 Neon free?

Gemini 4 Neon is available in the Gemini app to Google AI Pro and Ultra subscribers. API access is billed according to token usage, starting at $3 per million input tokens.

How much does Neon cost?

The API costs $3 per million input tokens and $15 per million output tokens. Cached input tokens cost $0.15 per million.

How can I access the model?

Users can access the model through the Gemini app with an eligible Pro or Ultra subscription. Developers can use it through Google AI Studio, Android Studio, and the Gemini API.

What is the difference between Neon and Carbon?

Neon focuses on advanced coding and analytical workflows with greater cost efficiency, while Carbon is designed for the most demanding and long-running tasks.

Does Neon replace Argon?

Neon replaces Argon in the Gemini app’s model selection for deeper analysis. Google has not announced a deprecation of Argon across its developer services.

Related Articles

Comments

No Comments Yet

Be the first to comment on this content.