Synthefy raises $6.5M seed and open-sources tabular foundation model Nori V1

On August 19, Synthefy announced a $6.5M seed round led by Wing VC and released Nori V1, an open-source (Apache 2.0) tabular foundation model. At just 30M parameters, it claims zero-shot performance matching Google's 1.6B-parameter TabFM and beating XGBoost on public regression benchmarks. Claims so far come mainly from official and reposted sources; independent third-party validation is not yet available, but the combination of zero-shot inference, interpretability, and commercial-friendly licensing makes it worth watching.

Confirmed

• Funding: $6.5M seed round led by Wing VC (m1, m2).
• Open source: Nori V1 code and weights under Apache 2.0, free for commercial use (m8).
• Zero-shot mechanism: Per @rohanpaulai, Nori uses in-context learning — fit merely stores Xtrain/ytrain as context while predict returns results in a single forward pass, requiring no feature engineering, training, or hyperparameter tuning; supports local execution (auto-pulled from Hugging Face) or API calls (m4, m6, m8, m9).
• Training data: trained entirely on synthetic data; supports predictive distributions (m9).
• Performance claims: Nori-30M matches Google's 1.6B-parameter TabFM on public regression benchmarks and beats XGBoost zero-shot (m2, m8).
• Interpretability: @rohanpaulai notes Nori is not a black box — predictions can be probed and feature contributions isolated to rebuild a transparent additive model; in a credit-card default case, 6 features recovered 95% of accuracy (m3).
• Multi-task reuse: Nori treats structured data (sensor readings, transactions, stock prices) as tokens, letting one pretrained model serve forecasting, pricing, fraud detection, risk analysis, and failure prediction (m5).
• API product: Nori Thinking is API-only, trading latency and cost for accuracy via more test-time compute; no downloadable weights (m7).

Why it matters

• Traditional tabular ML relies on per-dataset feature engineering and tuning; if Nori's zero-shot claims hold, it could lower barriers much like foundation models did in NLP (m6).
• 30M parameters rivaling a 1.6B competitor, plus distillation into interpretable small models, balances performance with audit/compliance needs (m2, m3).
• Caveat: performance comparisons and XGBoost-replacement claims come mainly from official and reposted sources; no independent verification appears in this batch of materials.

2026-08-19 ~ 2026-08-19 · 9 related posts

• Synthefy Raises $6.5M Seed to Launch Nori, Open-Source Tabular Foundation Model — WhatTheLJW · 2026-08-19
• [source] Nori V1: 30M Param Open-Source Model Beats XGBoost Zero-Shot — rohanpaul_ai · 2026-08-19
• [source] Nori-30M matches 1.6B parameter rivals in tabular data benchmarks — ai · 2026-08-19
• [source] Synthefy Releases Nori V1: Open Source Foundation Model Aiming to Replace XGBoost — rohanpaul_ai · 2026-08-19
• Nori handles multiple tabular tasks with one model — rohanpaul_ai · 2026-08-19
• Nori tabular model predicts without training weights — rohanpaul_ai · 2026-08-19
• Nori tabular model can be distilled into interpretable form — rohanpaul_ai · 2026-08-19
• Nori Thinking is API-only, trading compute for accuracy — rohanpaul_ai · 2026-08-19
• Nori: Open-Source Tabular Foundation Model for Fast, Training-Free Regression — rohanpaul_ai · 2026-08-19