shiyu-coder/Kronos: First Open-Source Foundation Model for Financial Candlestick Language — 11.9K Stars, AAAI 2026

GitHub·medium signal

Kronos treats K-line (candlestick) sequences as a domain-specific language for financial markets, applying NLP-style pre-training with a hierarchical tokenizer that converts continuous OHLCV data into discrete tokens followed by an autoregressive Transformer. Four model sizes from 4.1M to 499.2M parameters, trained across 45+ global exchanges. Accepted to AAAI 2026, this is the first open-source foundation model specifically for financial market forecasting — not generic time-series, but market-native language modeling.