Kronos is a family of open AI models trained on candlestick price data from markets, built to forecast prices and volatility and to generate realistic artificial price series.
In plain words
A candlestick record sums up a stretch of trading with its opening, highest, lowest and closing prices. In an approach inspired by language models, Kronos turns these records into tokens and trains a Transformer on them. Its makers used more than 12 billion records from 45 exchanges. Their paper reports that, with no extra training, it beats earlier time-series models at forecasting prices and volatility, and can produce realistic made-up price series. Several sizes are on Hugging Face, and a demo forecasts the BTC/USDT trading pair 24 hours ahead.
A quick example
Imagine a student testing an idea about one company's shares. They give the model a few months of hourly prices, ask what comes next, and compare its guesses with what really happened before trusting it with any decision.
Where you'll meet it
- Forecasting prices and volatility from market data.
- Making realistic artificial price data for testing.
Related
Sources
- Docsshiyu-coder/Kronos, Kronos (GitHub) · read 19 Oct 2026
- PaperKronos: A Foundation Model for the Language of Financial Markets, arXiv · read 19 Oct 2026
Checked by WhatIsAI against the sources above · 19 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.