Synthefy raises $6.5M to build foundation models for numbers, not words
What's the deal? SynthefyDealroom has a profile for this one. Try Dealroom → has raised $6.5 million in seed funding to expand a foundation-model platform tuned for numerical data rather than text. Wing Venture CapitalDealroom has a profile for this one. Try Dealroom → led the round, with participation from Haystack, Samsung Next, Canonical CryptoDealroom has a profile for this one. Try Dealroom →, and Lightscape PartnersDealroom has a profile for this one. Try Dealroom →. Angel investors from OpenAI, Microsoft, and Meta also backed the startup.
What's the endgame? Synthefy calls its approach "Structured Data Foundation Models," or SDFMs. Just as large language models learn word patterns from text, its models ingest numerical data to learn calculations and preserve relationships within time-series data and tables.
The product: The company also detailed Nori, its first open-source SDFM, released in August 2026. In testing, a 30-million-parameter version outperformed GoogleDealroom has a profile for this one. Try Dealroom →'s 1.6-billion-parameter TabFM model — despite being 2% its size. Nori has already been downloaded more than 600,000 times.
Why now? The models target workloads such as fraud detection, dynamic pricing, and demand forecasting — tasks enterprises usually handle with frameworks like LightGBMDealroom has a profile for this one. Try Dealroom → and XGBoostDealroom has a profile for this one. Try Dealroom →. Co-founder and chief executive officer Somi Agarwal said that approach forces teams to spend weeks on data preparation and tuning for each new problem.
"That work does not compound," Agarwal said. "But with Nori, a team can point the model at a new table or problem and get a strong prediction without training or tuning a new model for that dataset. That can bring the initial evaluation down from weeks or months to just minutes."
Enterprises can access the capabilities through an open model, a managed API, or by deploying within their own environments. Synthefy plans to monetise premium support and features atop its open foundation layer, and will use the funding to accelerate research, hire engineers, and build the next Nori.
The signal: The round sits in the top 8% of all-time US seed rounds for machine learning startups, a sign of investor appetite for applying the foundation-model playbook beyond text. If numerical data can be modelled the way language has been, the everyday enterprise work of pricing, forecasting, and risk analysis is the prize.
Read more: SiliconANGLE
Image credit: Synthefy
