As enterprises grapple with rising AI infrastructure costs, Lovelace today released benchmark results suggesting that better context, not bigger models, will be the next quantum leap in enterprise AI.
In a benchmark of 12 complex financial and business research tasks, a lightweight LLM powered by Lovelace achieved parity with Google’s Gemini Deep Research Max while operating at less than 1% of the cost.
The benchmark evaluated a range of complex financial and business questions, including company comparisons, acquisition scenarios, and investment analyses. Reports were judged based on factual accuracy, analytical rigor, use of evidence, and citation quality.
At the center of the benchmark is the YottaGraph, Lovelace’s flagship context engine. It provides AI agents with a continuously updated view of companies, people, markets, events, and relationships, connecting more than 60 million entities and billions of facts across multiple data sources in real time.
Gemini Deep Research had access to the public internet, while the Lovelace agent relied solely on YottaGraph and lacked the ability to search the internet. The benchmark was based on a simple question: Can an agent powered by a lightweight LLM hooked up to Lovelace’s YottaGraph and nothing else provide deep research-grade reports significantly cheaper and faster than a flagship Deep Research model? According to the results, it can.
“We are approaching the end of an era where every breakthrough in AI requires more compute, more power, and more spend,” said Andrew Moore, co-founder and CEO of Lovelace. “What we’re seeing is that context can do the heavy lifting. We spent the last two years building the context layer while everyone else was burning tokens.”
The findings arrive as enterprises increasingly confront the economic realities of deploying AI at scale. Many organizations are discovering that sophisticated AI workflows can consume large amounts of compute and generate significant infrastructure costs, creating pressure to find more efficient approaches.
“The future of AI isn’t about giving models more tokens. It’s about giving them more understanding,” Moore said.
Lovelace has detailed the benchmark on their blog, and made the methodology, evaluation framework, prompts, and sample reports publicly available on the Lovelace GitHub to support independent review and reproducibility.
