KRONOS · K-LINE FOUNDATION MODEL

Kronos Crypto Candle Forecast

Forecast the next candles of any USDT perpetual contract with Kronos, the open-source foundation model for financial K-lines — six intervals from 5 minutes to 1 day, with history and forecast on one chart.

How to use

Run a Kronos K-line analysis in three steps:

  1. Choose a coin

    Open the “Contract” dropdown and type a coin name (e.g. btc, eth) to search. Use ↑ ↓ to move and Enter to confirm. Popular coins are pinned at the top.

  2. Choose an interval

    Pick one of 5m, 15m, 30m, 1h, 4h or 1d. Shorter intervals focus on short-term moves; longer ones show the broader trend.

  3. Click Start analysis

    Click “Start analysis”. The latest 512 candles are sent to the Kronos model, and a few seconds later the chart shows history and forecast candles together, with the forecast change and range.

Interval
Sign in / sign up ↗

Sign in to use Kronos forecasts

Kronos

Choose a coin and an interval, then click “Start analysis”.

Forecasts reflect statistical patterns in past candles, are not guaranteed and are for research only. Not investment advice.

About the Kronos model

What is Kronos

Kronos is the first open-source, decoder-only foundation model for financial K-lines. A dedicated tokenizer quantizes OHLCV into hierarchical discrete tokens, and an autoregressive Transformer generates the following candles.

Training data

Per the authors, it was pre-trained on more than 12 billion K-line records from 45 global exchanges; the paper was accepted at AAAI 2026 (arXiv 2508.02739).

How this page uses it

After you pick a pair and interval, the server reads the latest 512 candles (the context length of Kronos-small / base) and asks Kronos for the open, high, low and close of the next candles.

Released Kronos models (source: official README)
ModelParametersContext length
Kronos-mini4.1M2048
Kronos-small24.7M512
Kronos-base102.3M512

Project: github.com/shiyu-coder/Kronos (MIT licence) · Paper: arXiv 2508.02739

Kronos is developed by its original authors; Coinpred only integrates it and is not affiliated with them. Model output is not investment advice.