New benchmark demonstrates enterprises can generate high-quality AI research locally, reducing the reliance on expensive cloud-based models while maintaining quality and keeping sensitive data in-house
PITTSBURGH, July 22, 2026 /PRNewswire/ — Lovelace today announced new benchmark results showing that enterprises can generate AI-powered research equivalent to Gemini Deep Research using open-source models running entirely on their own hardware.
The findings suggest organizations no longer need to send sensitive agent conversations to external AI providers or pay recurring cloud inference fees to produce sophisticated research and analysis.
In Lovelace’s latest benchmark, the company’s YottaGraph-context engine, running alongside a locally-hosted Gemma 4 model, achieved research-quality results virtually identical to Google’s Gemini Deep Research while reducing inference costs from approximately $7 per report to about one cent in electricity costs.
The benchmark builds on Lovelace’s earlier research, replacing the final cloud-based component with a locally hosted AI model. The result is an end-to-end AI research workflow in which the agent runs entirely within an organization’s own environment, giving enterprises greater control over cost, privacy, and compliance.
“For years, companies have assumed they needed the biggest models running in the cloud to get the best AI results,” said Andrew Moore, co-founder and CEO of Lovelace. “What this benchmark shows is that the real advantage isn’t the model; it’s the quality of the context you give it. With the right foundation, organizations can achieve exceptional results using open-source AI running entirely on their own infrastructure.”
For enterprises handling proprietary or regulated information, keeping AI workloads on-premises can simplify compliance while reducing reliance on third-party AI providers. At the same time, replacing recurring cloud inference fees with local compute can significantly lower the cost of deploying AI at scale.
The benchmark evaluated 18 complex investment banking research scenarios, including competitive analysis, strategic alternatives, investment memoranda, and commodities trading. Across the benchmark, Lovelace’s locally hosted system delivered research quality statistically equivalent to Gemini Deep Research while operating without cloud API-bound AI agents or external web search.
“Our industry has spent the last two years asking how to build bigger AI models,” said Moore. “The more important question is how to give AI better information. Once that hurdle is cleared, you don’t need the biggest model or the cloud to solve mission-critical business problems.”
The results represent another step toward Lovelace’s vision of enterprise AI that is more trusted, more affordable, more secure, and easier to govern. By combining its YottaGraph context engine with locally deployed open-source models, organizations can keep sensitive information behind their firewall while dramatically reducing the ongoing cost of AI-powered research.
The complete benchmark methodology and technical analysis are available on the Lovelace blog.
For more information, visit lovelace.ai.
About Lovelace
Lovelace is the only provider of enterprise-scale context engines capable of analyzing trillions of real-time data points to create knowledge graphs that are usable by autonomous agents at the speed, scale, and accuracy required for mission-critical analysis. Lovelace’s context engine platform, Elemental, uniquely integrates data ingestion, entity resolution, and graph building into a single pipeline that empower agentic deployments, exponentially amplifying the investigative power for complex queries. With its proprietary ground-breaking YottaGraph, Lovelace provides enterprises with real-time context, enabling agents to understand the impact of global intelligence on enterprise data for unmatched insights with millisecond precision. Founded in 2023 by Andrew Moore, former head of Google Cloud AI and dean of Carnegie Mellon’s School of Computer Science, Lovelace currently works with some of the largest public and private enterprises in the world.
View original content to download multimedia:https://www.prnewswire.com/news-releases/lovelace-shows-that-local-models-can-match-the-best-cloud-based-agentic-systems-for-frontier-ai-302831544.html
SOURCE Lovelace AI
