Google Unveils Gemini 4 Neon: A Revolutionary Model for Complex Workflows
In an innovative leap forward, Google has introduced Gemini 4 Neon, a cutting-edge model meticulously crafted to proficiently manage intricate software engineering, scientific inquiries, and enterprise operations.
This groundbreaking variant promises to deliver performance on par with the more robust Gemini 4 Carbon model but does so at an impressive threefold increase in speed while reducing costs to 60 percent.
Engineered for sustained, multi-stage operations, the Neon model excels in debugging large codebases, scrutinizing extensive datasets, and orchestrating the efforts of multiple agents to achieve sophisticated project goals.
Preliminary internal assessments conducted by Google suggest that Neon not only matched but at times surpassed Carbon in overall quality, achieving these feats with remarkable speed.
Notably, this model has recorded its lowest fabrication rate yet during internal tests conducted within Google’s Antigravity framework.
Neon has achieved a score of 61 on the Artificial Analysis Intelligence Index at the highest “xhigh” cognitive level, securing third place overall, according to Google’s findings.
Additionally, it ranked second on both the Vals Index and the newly established Agent Arena Index, which evaluates the model’s adaptability to user adjustments during task execution.
Accelerated Coding and Research Capabilities
A significant emphasis of this release is on the realm of software engineering. Neon has excelled, attaining the top position on the Artificial Analysis Coding Agent Index with a score of 58.6, alongside a commendable 65.2 percent on Terminal-Bench 4.1.
Within Google’s Antigravity environment, the model achieved a notable 39 percent on SWE-sweep—more than a threefold improvement over its prior 12.8 percent performance on the mini-SWE harness.
These capabilities are intricately designed to empower developers in navigating multiple file systems, troubleshooting intricate repositories, and implementing precise modifications without the need for redundant processing of information.
Furthermore, Neon functions as an orchestrator within Google Antigravity, adeptly directing parallel Hydrogen subagents to tackle various components of a single task.
Throughout prolonged sessions, it communicates progress and maintains robust performance even as workflows grow increasingly intricate.
Beyond its coding endeavors, Google envisions Neon as a vital instrument for enterprise analysis and scientific research.
Its expansive 1-million-token context window equips it to manage substantial volumes of data across spreadsheets, PDFs, charts, audio files, and videos.
In various assessments, the model has been recognized with top rankings, notably in the Legal Research Bench, Tax Agent Bench, and Zapier’s AutomationBench.
Additionally, it scored an impressive 70.5 percent on Terminal-Bench Science and 84 percent on RiemannBench, an evaluation focused on mathematical reasoning.
Cost-Efficiency and Enhanced Security Features
In the realm of economics and cybersecurity, Google is promoting Neon’s competitive pricing structure and advanced security features.
The model is available at a price of $3 per million input tokens and $15 per million output tokens, with cached input tokens priced at an unprecedented $0.15 per million, indicating a staggering 95 percent discount.
At its standard high cognitive level, Neon has demonstrated a 32 percent reduction in task costs compared to Carbon.
Moreover, Google revealed cost reductions of approximately 50 percent to 80 percent in comparison with leading rival models across several evaluations.
For cybersecurity applications, a specialized iteration of Neon is devised to assist defenders in pinpointing vulnerabilities, assessing risks, and rectifying software inadequacies.
It achieved a remarkable 90.5 percent on SEC-bench Pro and 76 percent on the challenging tier of AutoBaxBench, as documented by Google.
Access to this specialized cybersecurity version will be initially restricted to trusted defenders participating in Google’s Fairwind Program.
Gemini 4 Neon is set to be deployed through the Gemini API in Google AI Studio and Gemini for business environments.
Subscribers to Google AI Pro and Ultra will also be able to access the model via the Gemini app and Antigravity, further broadening its reach.

The introduction of Neon enriches Google’s Gemini 4 suite, strategically situated between the high-speed Hydrogen model and the more potent Carbon model, catering to the needs of demanding, high-frequency workloads.
Source link: Interestingengineering.com.



