Google Unveils Gemini 4 Neon to Balance Performance and AI Costs
Google introduced the Gemini 4 Neon AI model with a 1 million token context window, balancing reasoning power and API efficiency for developers.
Google has introduced Gemini 4 Neon, a new artificial intelligence model designed to balance reasoning performance with operational costs, offering a 1 million token context window.
Google has introduced Gemini 4 Neon, a new artificial intelligence model designed to balance processing efficiency with reasoning performance. According to details shared by Webrazzi, the model is built for complex software engineering, enterprise workflows, and scientific research. It sits within the Gemini family between the flagship Carbon and the everyday Hydrogen model. While the architecture accepts text, image, audio, and video inputs, it does not produce direct image or audio outputs.
According to ShiftDelete.Net, the release is now available in the Gemini app for Google AI Pro and Google AI Ultra subscribers.
Alongside the launch, Google revised its consumer model catalog, deprecating Argon and removing Gemini 3.1 Pro. Gemini 3.5 Flash-Lite now serves as an automated routing option for paid subscribers, helping users balance speed and operational expenses.
Architecture and Reasoning Levels of Gemini 4 Neon
Developed from the foundation of Gemini 4 Carbon, Gemini 4 Neon provides a unified context window of 1 million tokens. The system supports up to 64,000 tokens in standard synchronous output, expanding to 1 million tokens in extended output mode. Developers can select from four reasoning tiers: low, medium, high, and extra high. The default high tier addresses general engineering tasks, while extra high targets intensive scientific analysis.
Expanding Carbon's three reasoning levels to four gives developers tighter control over task-specific compute costs.
The model also assists office and data operations, reading spreadsheets, PDFs, and photos with greater accuracy while processing enterprise data from Gmail and Docs. Built-in live code execution and Google Search verification support persistent artificial intelligence agents. The model can operate independently in Google Antigravity or coordinate multiple Gemini 4 Hydrogen sub-agents simultaneously.
Benchmark Results Across Code and Science
Google's published model card shows Gemini 4 Neon scored 61 points on the Artificial Analysis Intelligence Index v5.0.1, placing third overall. It ranked ahead of Gemini 4 Carbon (60), Claude Opus 5.5 (59), and GPT-6.2 Sol (57). Neon also led the Vals Index with 70.9 percent over Claude Opus 5.5 (70.1 percent) and GPT-6.2 Sol (66.1 percent). In software evaluations, Neon reached 65.4 percent on CursorBench 4.1, surpassing Claude Opus 5.5 (58.1 percent).
On Terminal-Bench 4.1, Neon scored 65.2 percent just behind Claude Opus 5.5 (65.8 percent), while Claude Opus 5.5 led FrontierCode 2.0 with 47.3 percent over Neon's 46.2 percent.
In scientific testing, Neon scored 84 percent on RiemannBench, 70.5 percent on Terminal-Bench Science, and 71.3 percent on HLE-Diamond. It topped rivals on Zapier AutomationBench with 56.3 percent, though GPT-6.2 Sol led Chartography visual analysis with 75.1 percent to Neon's 73.5 percent. In Google Antigravity, Neon recorded 39 percent on the SWE-sweep benchmark and took second place on Agent Arena with 1336 points across more than 4,200 sessions.
API Pricing and Enterprise Availability
Standard API pricing for Gemini 4 Neon is $3 per 1 million input tokens and $15 per 1 million output tokens, dropping to $0.15 per 1 million tokens for cached input. Google notes that running at the default high reasoning level reduces per-task costs by 32 percent compared to Carbon. Carbon's promotional pricing ends Nov. 30, 2026, rising Dec. 1, 2026, to $4 for inputs and $20 for outputs per 1 million tokens.
Following that increase, Neon will remain 25 percent cheaper than Carbon in standard list token pricing.
Neon also supports defensive cybersecurity workflows, including secure coding, vulnerability detection, and patch creation, with a specialized edition offered to researchers under Google's Fairwind Program. The model is accessible via the Gemini API, Google AI Studio, and enterprise Gemini plans. Local billing terms and official pricing in Turkish lira (TL) for Türkiye have not been confirmed.
In TeknoRest's editorial view, Google's introduction of Gemini 4 Neon directly addresses the challenge of balancing reasoning capability with operational expenses. The $3 entry price per million input tokens and the 32 percent cost reduction at the default high setting provide tangible relief for engineering teams running continuous workloads. While a 65.4 percent score on CursorBench 4.1 demonstrates reliable coding capabilities, the 39 percent outcome on SWE-sweep indicates that autonomous software engineering still requires human oversight. Although the lack of direct audio and image generation might appear restrictive, focusing purely on text and code is a practical choice that conserves compute resources for enterprise infrastructure.
Frequently Asked Questions (FAQ)
What input and output formats does Gemini 4 Neon support?
Gemini 4 Neon accepts text, image, audio, and video inputs. However, it does not produce direct image or audio outputs, generating only text and code formats.
What is the API pricing for Gemini 4 Neon?
Standard API pricing is set at $3 per 1 million input tokens and $15 per 1 million output tokens. For cached input prompts, the rate drops to $0.15 per 1 million tokens.
Who can access Gemini 4 Neon in the Gemini app?
In the Gemini app, the model is available to subscribers of Google AI Pro and Google AI Ultra.
- →Google Launches Gemini Tasks for Free Across Six Countries
- →Google Adds Desktop Homepage Sidebar for Google AI Search
- ◆Explore Category & Analysis: Technology News & Analysis Hub
- ↗Official Reference & Documentation: Google DeepMind Official Model Documentation (deepmind.google)