AI × Markets

AI just learned the language of candlesticks.

Kronos is an open-source foundation model built specifically for financial K-lines — the open, high, low, close and volume behind every candle. Instead of reading a chart as an image, it turns market data into tokens and learns the sequence like a language model.

🌍 45 global exchanges 🧠 12B+ K-line records 🏆 AAAI 2026
📈 Price forecasting🌊 Volatility🧪 Synthetic data 🎯 Zero-shot🔧 Fine-tunable🆓 MIT licensed
The 60-second version

It treats candles like words.

A specialised tokenizer converts continuous OHLCV market data into hierarchical discrete tokens. A decoder-only Transformer then learns the patterns between those tokens and generates possible future K-line sequences.

OHLCV candles→tokenizer→market tokens→Transformer→forecast
4.1MKronos-mini parameters
24.7MKronos-small parameters
102.3MKronos-base parameters

Reality check: this is a research model, not a magic signal generator. Forecasts are probabilistic, markets change regime, and a backtest is not a promise of future returns. Use it to research — not to outsource risk.

Try Kronos yourself

1 See the live BTC demo

The official demo visualises a 24-hour BTC/USDT forecast. Start here if you want to see the output before installing anything.

→ Open the official Kronos demo

2 Clone the open-source repo

You need Python 3.10+ and Git.

git clone https://github.com/shiyu-coder/Kronos.git cd Kronos

3 Create an environment + install

python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt
Windows activation: .venv\Scripts\activate

4 Run the forecasting example

The example loads a pretrained Kronos model, forecasts the next window and renders predicted candles against the ground truth.

python examples/prediction_example.py
The pretrained tokenizer and model are downloaded from Hugging Face on first use.

5 Bring your own market data

Feed KronosPredictor a pandas DataFrame with open, high, low and close. Volume and amount are optional. The repo also includes a full fine-tuning pipeline for adapting the tokenizer and predictor to your own domain.

Go deeper
⭐ Get the code — shiyu-coder/Kronos
Models, examples, batch prediction and the fine-tuning pipeline. Star the original project if it helps you.
Open GitHub →
🤗 Download the pretrained models
Kronos-mini, Kronos-small, Kronos-base and their tokenizers on Hugging Face.
Open the model collection →
📄 Read the research paper
Architecture, pretraining corpus, benchmarks and ablations — the evidence behind the claims.
Open the paper →

More AI × markets, without the hype.

Follow @cloud9.markets for open-source tools, honest tests and the next build worth trying.

Follow @cloud9.markets →