Exclusive: Synthefy raises $6.5M for its number-crunching models trained on numerical data instead of words

Synthefy Inc., a startup specializing in numerical data models, has secured $6.5 million in seed funding to expand its foundation-model platform, tailored for numerical data analysis. This investment was led by Wing Venture Capital and joined by Haystack, Samsung Next, Canonical Crypto, and Lightscape. Notable angel investors included OpenAI Group PBC, Microsoft Corp., and Meta Platforms Inc., indicating strong industry support.

Synthefy pioneered "Structured Data Foundation Models" or SDFMs, which learn from numerical data, similar to how large language models (LLMs) process textual data. These models excel in preserving intricate relationships within time-series data and tables, enabling faster and more accurate calculations compared to traditional machine learning models.

Synthefy recently released its first open-source SDFM, Nori, which despite its lightweight architecture, outperformed Google's 1.6-billion parameter TabFM model in a benchmark test, even with its "Thinking" capabilities enabled. Nori offers significant performance advantages when pre-trained on specific tasks, allowing enterprises to quickly train and deploy specialized models for applications like fraud detection, pricing optimization, and demand forecasting.

The company believes it can monetize its open-source models through premium support, managed API usage, and private deployments, leveraging its strong enterprise traction and community adoption, with over 600,000 downloads of Nori in just weeks.