Kronos

Kronos is an open-source foundation model for financial candlestick (K-line) data. The authors say it was trained on data from more than 45 exchanges worldwide, and the paper was accepted at AAAI 2026. It first turns continuous open, high, low, close and volume data into discrete tokens, then has a large model learn this "language of the market" for a range of quantitative forecasting tasks. The repository releases mini, small and base models, while large isn't open-sourced, and a few lines of Python load a model, take historical candles and return a forecast. The README also includes fine-tuning scripts and an A-share example that fine-tunes on your own data, generates prediction signals and backtests a stock-picking strategy. The authors host a live demo showing a 24-hour forecast for the BTC/USDT pair. What sets it apart is that it doesn't treat prices as ordinary time series; it's designed around the high noise of financial data.

Who it's for:researchers and developers in quantitative finance who want to try a model built for financial time series. Readers who don't write code won't have a use for it.

Before you start: we list it as a research model, and its forecasts are not investment advice. Markets are highly uncertain, so don't trade directly on its predictions. The README itself says the fine-tuning example is a demonstration, not a production-ready trading system. The live demo uses a cryptocurrency pair, and we don't recommend trading crypto on the strength of it. It needs Python 3.10 or later; the model weights are hosted on Hugging Face and have their own licence, which you'll need to check. The repository code is MIT-licensed. We didn't see any feature for placing orders on an exchange in the documentation we read.

Based on the project's public documentation. Not tested by AI Inspiration World.