News

Google releases Gemini 4 Neon AI model for low-cost, complex coding and analysis

Published: | Posted By:

The push for "mid-tier" efficiency in generative AI has reached a new milestone with the launch of Gemini 4 Neon, a model designed to deliver flagship-level coding and analytical performance without the associated cost of the largest proprietary models. Released on November 18, 2026, Neon is positioned as a workhorse for complex software engineering, data analysis, and multi-step workflows, effectively bridging the gap between lightweight daily assistants and heavy-duty research tools.

This release significantly reshuffles Google's AI hierarchy, which now follows a periodic table naming convention. Gemini 4 Neon sits between the small, fast Gemini 4 Hydrogen and the heavy-duty flagship Gemini 4 Carbon. To make room for this new mid-range option, Google has retired Gemini 3.1 Pro and abruptly pulled Gemini 4 Argon from the model picker in the Gemini app, just seven weeks after Argon's own launch. Neon inherits a one-million-token context window from Carbon and allows developers to toggle between four reasoning levels: low, medium, high, and xhigh.

While positioned as a mid-tier model, Neon's performance benchmarks suggest it can compete with the industry's top offerings. At its xhigh setting, Neon scored 61 on the Artificial Analysis Intelligence Index, placing it third overall behind only OpenAI’s GPT-6.2 Astra and Anthropic’s Claude Fable 5.2. It also claimed the top spot at launch on the Artificial Analysis Coding Agent Index with a score of 58.6. However, this speed and intelligence come with a slight trade-off in reliability; Neon recorded a 13.9% factual hallucination rate at the high setting, which is higher than both Carbon's 11.4% and the retired Argon's 4.3%, though it remains significantly lower than Claude Opus 5.5's 50.3%.

These technical capabilities translate into specific enterprise utilities where human oversight can verify the output. Google and early users highlight tasks such as tracing unexpected charges across billing records, identifying flaws in unfamiliar software architectures, and testing research hypotheses. In one demonstration of its analytical muscle, Neon agents reconstructed a single image across 100,000 spreadsheet cells using over 500,000 formulas to map depth and lighting without using a script or embedding the original picture. This allows a human collaborator to open the resulting sheet and manually inspect the formulas, mitigating the "black box" nature of AI-generated work.

The economic appeal of Neon is central to its rollout, though the current pricing presents a temporary quirk. Neon is priced at $3 per million input tokens and $15 per million output tokens. While this is 25% cheaper than the standard rates for Anthropic's Opus 5.5, it is currently more expensive than Gemini 4 Carbon, which is running on a promotional rate of $2 per million input and $10 per million output through November 30. Once Carbon moves to its standard $4 and $20 pricing on December 1, Neon becomes the more cost-effective option for mid-tier workloads.

For developers and enterprise users, Neon is available via the Gemini API in Google AI Studio and Android Studio, as well as for Google AI Pro and Ultra subscribers in the Gemini app. Within the Google Antigravity development platform, Neon can be utilized in "Orchestrator" mode, where it acts as a lead architect planning codebase changes while delegating simpler tasks to Hydrogen sub-agents. It is worth noting that some users in India on telecom-bundled Pro plans have reported entitlement errors preventing access to Gemini 4 models within Antigravity.

The introduction of a high-performance mid-tier model reflects a broader industry shift toward "supervised autonomy." While Neon can handle multi-file debugging and refactoring, industry analysts suggest that the complexity of enterprise codebases remains a barrier to fully autonomous deployment. The trend is moving toward a tiered approach, where enterprises select models based on the specific economics and requirements of a task, using models like Neon to reduce costs while maintaining a level of quality that allows human developers to retain final accountability.

Read more from Ninjalane
Hardware Reviews, News and More...