Skip to content
Back to the Guavy Wire
Stocks

Noah Labs Triples Accuracy with Google Cloud's Gemma 4 Model

Instruments
GOOGL
Share

Noah Labs, a Palo Alto-based company specializing in AI tools for defense contractors, government agencies, and enterprises, has achieved significant improvements in code modernization by leveraging Google Cloud. The company's Sentinel system, powered by Google DeepMind's Gemma 4 31B model, delivers 3x higher accuracy per dollar compared to alternative models, while reducing evaluation spend by 3.1x to 7.5x across teams.

The key to Sentinel's success lies in its structured orchestration framework, which manages five critical stages of the engineering workflow: context retrieval, tool invocation, task orchestration, automated verification, and structured reviews. This approach ensures that code modifications are thoroughly checked before human review, enhancing efficiency and reliability. Murat Can Isik, Co-founder and CEO of Noah Labs, emphasizes the importance of speed and cost-effectiveness in the development cycle, stating that faster generation and lower GPU costs make iterations more responsive and accessible for teams.

Noah Labs evaluated six different open AI models within the Sentinel harness, focusing on real-world performance rather than synthetic leaderboard rankings. The Gemma 4 31B model stood out, delivering 65.6% faster response speeds and outperforming larger models in accuracy benchmarks. This model achieved 97.6% coding accuracy, 93.8% on MATH-500, and 97.0% on GSM8K, while maintaining a 168.4 useful-work-per-second score.

The economic and operational benefits of using Gemma 4 31B on Google Cloud are substantial. Noah Labs reduced its evaluation compute spend significantly, proving that lean hardware configurations can outperform heavy models without compromising engineering precision. This advancement allows software teams to modernize complex legacy codebases more efficiently, turning them into understandable, testable, and maintainable systems. Looking ahead, Noah Labs plans to integrate deeper automated verification features and explore further deployment options with Gemini models.

More on Stocks