Key Highlights
- Google introduced Gemini 4 Neon on November 18, 2026.
- Gemini 4 Neon scored 61 on the Artificial Analysis Intelligence Index at xhigh thinking level.
- The model ranked second on the Vals Index with a score of 70.9%.
- Gemini 4 Neon features a 1 million-token context window.
- API pricing starts at $3 per million input tokens and $15 per million output tokens.
- Cached input tokens are priced at $0.15 per million tokens.
- Google reported 3x faster execution and 60% lower cost compared with Gemini 4 Carbon in early evaluations.
- The model achieved 39.0% on SWE-sweep using Google’s Antigravity harness.
- Gemini 4 Neon scored 90.5% on SEC-bench Pro.
- Gemini 4 Neon for cyber will roll out to eligible defenders through the Fairwind Program.
Google has introduced Gemini 4 Neon, a new artificial intelligence model developed by Google DeepMind, targeting complex software engineering, enterprise knowledge work, scientific research and multi-step agentic workflows. Announced on November 18, 2026, the model is designed to deliver frontier-level intelligence with faster execution and lower operating costs compared with larger AI models.
According to Google, Gemini 4 Neon secured a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, ranking third overall. The company said the model also achieved second place on the Vals Index with a score of 70.9%, placing it among the leading AI models in its performance category.
The new model is designed to balance intelligence, speed and cost efficiency, particularly for businesses and developers handling complex tasks that require multiple steps, external tools and large volumes of information.
Gemini 4 Neon Pricing and Context Window
Google has positioned Gemini 4 Neon as a cost-efficient alternative for demanding AI workloads. The model comes with a context window of up to 1 million tokens, allowing it to process large documents, code repositories, spreadsheets and other information within a single context.
The API pricing structure is as follows:
- Input tokens: $3 per million tokens.
- Cached input tokens: $0.15 per million tokens.
- Output tokens: $15 per million tokens.
Google stated that cached input tokens receive a 95% discount compared with standard input pricing. The company also reported that Gemini 4 Neon matched or exceeded Gemini 4 Carbon in overall quality during early internal evaluations while operating at three times the speed and 60% of the cost.
Strong Performance Across Coding and Software Engineering Benchmarks
Gemini 4 Neon has been developed to support production-level software engineering, including multi-file debugging, code refactoring, repository exploration and long-running agentic tasks.
According to Google, the model ranked first on the Artificial Analysis Coding Agent Index with a score of 58.6. It also recorded 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0.
On SWE-sweep, Gemini 4 Neon achieved 39.0% using Google’s Antigravity evaluation environment, compared with 12.8% on the mini-SWE harness.
The model is also designed to operate as an orchestrator within Google Antigravity, coordinating parallel Hydrogen subagents to complete complex tasks. Google said Neon reduces unnecessary file reads, maintains progress updates and manages context during extended sessions.
Performance in Enterprise and Knowledge Work
Google said Gemini 4 Neon can process information from spreadsheets, PDFs, charts, audio and video within its 1 million-token context window. The model is designed to convert unstructured information into structured reports, analytical outputs and business deliverables.
On the Vals Index, it achieved a score of 70.9%, ranking second. It also recorded 56.3% on AutomationBench and 75.8% on Agents’ Last Exam v1.1.
The company further reported that the model achieved first-place results on the Legal Research Bench and Tax Agent Bench.
In addition, Gemini 4 Neon recorded a score of 73.5% on Chartography, which evaluates the ability of AI models to interpret complex professional charts.
Advances in Scientific and Mathematical Reasoning
Gemini 4 Neon also demonstrated performance across scientific research, mathematics and technical problem-solving benchmarks.
Google reported that the model achieved 84.0% on RiemannBench, 70.5% on Terminal-Bench Science 0.1 and 71.3% on HLE-Diamond.
These results highlight the model’s intended capabilities in quantitative analysis, mathematical reasoning and technical research workflows.
Gemini 4 Neon for Cybersecurity
Alongside general-purpose capabilities, Google has introduced a specialised Gemini 4 Neon for cyber version aimed at cybersecurity professionals and trusted defenders.
The cybersecurity-focused model is designed to assist with identifying software vulnerabilities, validating security risks and developing patches to strengthen existing code.
Google reported that Neon achieved 90.5% on SEC-bench Pro and 70.7% on CWE-bench v2. It also recorded 51.26% on WebExploitBench v1.2.
On AutoBaxBench, the model achieved a 76% score in the hard challenge category, according to the company.
Gemini 4 Neon for cyber will initially be made available to eligible cybersecurity professionals through Google’s Fairwind Program.
Safety and Reliability Improvements
Google said Gemini 4 Neon incorporates safety and security improvements developed for its Gemini 4 model family. These include protections against malicious prompt injections, unauthorised actions and potential misuse in sensitive areas such as cybersecurity and biotechnology.
The company also highlighted improvements in instruction following and model alignment, with a focus on reducing fabricated outputs and preventing AI agents from exceeding user instructions.
According to Google’s internal evaluations, Gemini 4 Neon achieved a 99.2% pass rate on its agent honesty evaluation and recorded a 0.09% misaligned action rate on its agentic safety evaluation.
Google also reported that the model resisted all attacks in the Gray Swan indirect prompt injection benchmark when operating with real-time safeguards at its highest thinking level.
Gemini 4 Neon Availability
Google has started rolling out Gemini 4 Neon across several of its AI platforms.
Developers can access the model through the Gemini API in Google AI Studio, while enterprise customers can use it through Gemini for business and Google Antigravity.
Google AI Pro and Ultra subscribers can access Gemini 4 Neon through Google Antigravity and the Gemini app. The company is also making the model available through its enterprise offerings.
The cybersecurity version will be distributed separately to eligible members of the Fairwind Program.
Positioning Within the Gemini 4 Model Family
Gemini 4 Neon occupies a middle position within Google’s Gemini 4 model family. While Gemini 4 Carbon is designed for demanding, long-running projects and Gemini 4 Hydrogen focuses on faster everyday execution, Neon is positioned to balance advanced reasoning with speed and cost efficiency.
With its emphasis on software development, enterprise automation, scientific analysis and cybersecurity, Gemini 4 Neon represents Google’s effort to make advanced AI capabilities more practical for frequent and complex professional workloads.



